Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 1968e438

History | View | Annotate | Download (311 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file h264.c
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264.h"
32
#include "h264data.h"
33
#include "h264_parser.h"
34
#include "golomb.h"
35
#include "rectangle.h"
36
#include "vdpau_internal.h"
37

    
38
#include "cabac.h"
39
#ifdef ARCH_X86
40
#include "x86/h264_i386.h"
41
#endif
42

    
43
//#undef NDEBUG
44
#include <assert.h>
45

    
46
/**
47
 * Value of Picture.reference when Picture is not a reference picture, but
48
 * is held for delayed output.
49
 */
50
#define DELAYED_PIC_REF 4
51

    
52
static VLC coeff_token_vlc[4];
53
static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
54
static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
55

    
56
static VLC chroma_dc_coeff_token_vlc;
57
static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
58
static const int chroma_dc_coeff_token_vlc_table_size = 256;
59

    
60
static VLC total_zeros_vlc[15];
61
static VLC_TYPE total_zeros_vlc_tables[15][512][2];
62
static const int total_zeros_vlc_tables_size = 512;
63

    
64
static VLC chroma_dc_total_zeros_vlc[3];
65
static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
66
static const int chroma_dc_total_zeros_vlc_tables_size = 8;
67

    
68
static VLC run_vlc[6];
69
static VLC_TYPE run_vlc_tables[6][8][2];
70
static const int run_vlc_tables_size = 8;
71

    
72
static VLC run7_vlc;
73
static VLC_TYPE run7_vlc_table[96][2];
74
static const int run7_vlc_table_size = 96;
75

    
76
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
77
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
78
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
79
static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
80
static Picture * remove_long(H264Context *h, int i, int ref_mask);
81

    
82
static av_always_inline uint32_t pack16to32(int a, int b){
83
#ifdef WORDS_BIGENDIAN
84
   return (b&0xFFFF) + (a<<16);
85
#else
86
   return (a&0xFFFF) + (b<<16);
87
#endif
88
}
89

    
90
static const uint8_t rem6[52]={
91
0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
92
};
93

    
94
static const uint8_t div6[52]={
95
0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
96
};
97

    
98
static const int left_block_options[4][8]={
99
    {0,1,2,3,7,10,8,11},
100
    {2,2,3,3,8,11,8,11},
101
    {0,0,1,1,7,10,7,10},
102
    {0,2,0,2,7,10,7,10}
103
};
104

    
105
#define LEVEL_TAB_BITS 8
106
static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
107

    
108
static void fill_caches(H264Context *h, int mb_type, int for_deblock){
109
    MpegEncContext * const s = &h->s;
110
    const int mb_xy= h->mb_xy;
111
    int topleft_xy, top_xy, topright_xy, left_xy[2];
112
    int topleft_type, top_type, topright_type, left_type[2];
113
    const int * left_block;
114
    int topleft_partition= -1;
115
    int i;
116

    
117
    top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
118

    
119
    //FIXME deblocking could skip the intra and nnz parts.
120
    if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
121
        return;
122

    
123
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
124
     * stuff, I can't imagine that these complex rules are worth it. */
125

    
126
    topleft_xy = top_xy - 1;
127
    topright_xy= top_xy + 1;
128
    left_xy[1] = left_xy[0] = mb_xy-1;
129
    left_block = left_block_options[0];
130
    if(FRAME_MBAFF){
131
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
132
        const int top_pair_xy      = pair_xy     - s->mb_stride;
133
        const int topleft_pair_xy  = top_pair_xy - 1;
134
        const int topright_pair_xy = top_pair_xy + 1;
135
        const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
136
        const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
137
        const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
138
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
139
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
140
        const int bottom = (s->mb_y & 1);
141
        tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
142

    
143
        if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
144
            top_xy -= s->mb_stride;
145
        }
146
        if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
147
            topleft_xy -= s->mb_stride;
148
        } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
149
            topleft_xy += s->mb_stride;
150
            // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
151
            topleft_partition = 0;
152
        }
153
        if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
154
            topright_xy -= s->mb_stride;
155
        }
156
        if (left_mb_field_flag != curr_mb_field_flag) {
157
            left_xy[1] = left_xy[0] = pair_xy - 1;
158
            if (curr_mb_field_flag) {
159
                left_xy[1] += s->mb_stride;
160
                left_block = left_block_options[3];
161
            } else {
162
                left_block= left_block_options[2 - bottom];
163
            }
164
        }
165
    }
166

    
167
    h->top_mb_xy = top_xy;
168
    h->left_mb_xy[0] = left_xy[0];
169
    h->left_mb_xy[1] = left_xy[1];
170
    if(for_deblock){
171
        topleft_type = 0;
172
        topright_type = 0;
173
        top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
174
        left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
175
        left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
176

    
177
        if(MB_MBAFF && !IS_INTRA(mb_type)){
178
            int list;
179
            for(list=0; list<h->list_count; list++){
180
                //These values where changed for ease of performing MC, we need to change them back
181
                //FIXME maybe we can make MC and loop filter use the same values or prevent
182
                //the MC code from changing ref_cache and rather use a temporary array.
183
                if(USES_LIST(mb_type,list)){
184
                    int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
185
                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
186
                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
187
                    ref += h->b8_stride;
188
                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
189
                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
190
                }
191
            }
192
        }
193
    }else{
194
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
195
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
196
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
197
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
198
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
199

    
200
    if(IS_INTRA(mb_type)){
201
        int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
202
        h->topleft_samples_available=
203
        h->top_samples_available=
204
        h->left_samples_available= 0xFFFF;
205
        h->topright_samples_available= 0xEEEA;
206

    
207
        if(!(top_type & type_mask)){
208
            h->topleft_samples_available= 0xB3FF;
209
            h->top_samples_available= 0x33FF;
210
            h->topright_samples_available= 0x26EA;
211
        }
212
        if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
213
            if(IS_INTERLACED(mb_type)){
214
                if(!(left_type[0] & type_mask)){
215
                    h->topleft_samples_available&= 0xDFFF;
216
                    h->left_samples_available&= 0x5FFF;
217
                }
218
                if(!(left_type[1] & type_mask)){
219
                    h->topleft_samples_available&= 0xFF5F;
220
                    h->left_samples_available&= 0xFF5F;
221
                }
222
            }else{
223
                int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
224
                                ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
225
                assert(left_xy[0] == left_xy[1]);
226
                if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
227
                    h->topleft_samples_available&= 0xDF5F;
228
                    h->left_samples_available&= 0x5F5F;
229
                }
230
            }
231
        }else{
232
            if(!(left_type[0] & type_mask)){
233
                h->topleft_samples_available&= 0xDF5F;
234
                h->left_samples_available&= 0x5F5F;
235
            }
236
        }
237

    
238
        if(!(topleft_type & type_mask))
239
            h->topleft_samples_available&= 0x7FFF;
240

    
241
        if(!(topright_type & type_mask))
242
            h->topright_samples_available&= 0xFBFF;
243

    
244
        if(IS_INTRA4x4(mb_type)){
245
            if(IS_INTRA4x4(top_type)){
246
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
247
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
248
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
249
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
250
            }else{
251
                int pred;
252
                if(!(top_type & type_mask))
253
                    pred= -1;
254
                else{
255
                    pred= 2;
256
                }
257
                h->intra4x4_pred_mode_cache[4+8*0]=
258
                h->intra4x4_pred_mode_cache[5+8*0]=
259
                h->intra4x4_pred_mode_cache[6+8*0]=
260
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
261
            }
262
            for(i=0; i<2; i++){
263
                if(IS_INTRA4x4(left_type[i])){
264
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
265
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
266
                }else{
267
                    int pred;
268
                    if(!(left_type[i] & type_mask))
269
                        pred= -1;
270
                    else{
271
                        pred= 2;
272
                    }
273
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
274
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
275
                }
276
            }
277
        }
278
    }
279
    }
280

    
281

    
282
/*
283
0 . T T. T T T T
284
1 L . .L . . . .
285
2 L . .L . . . .
286
3 . T TL . . . .
287
4 L . .L . . . .
288
5 L . .. . . . .
289
*/
290
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
291
    if(top_type){
292
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
293
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
294
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
295
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
296

    
297
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
298
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
299

    
300
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
301
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
302

    
303
    }else{
304
        h->non_zero_count_cache[4+8*0]=
305
        h->non_zero_count_cache[5+8*0]=
306
        h->non_zero_count_cache[6+8*0]=
307
        h->non_zero_count_cache[7+8*0]=
308

    
309
        h->non_zero_count_cache[1+8*0]=
310
        h->non_zero_count_cache[2+8*0]=
311

    
312
        h->non_zero_count_cache[1+8*3]=
313
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
314

    
315
    }
316

    
317
    for (i=0; i<2; i++) {
318
        if(left_type[i]){
319
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
320
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
321
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
322
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
323
        }else{
324
            h->non_zero_count_cache[3+8*1 + 2*8*i]=
325
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
326
            h->non_zero_count_cache[0+8*1 +   8*i]=
327
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
328
        }
329
    }
330

    
331
    if( h->pps.cabac ) {
332
        // top_cbp
333
        if(top_type) {
334
            h->top_cbp = h->cbp_table[top_xy];
335
        } else if(IS_INTRA(mb_type)) {
336
            h->top_cbp = 0x1C0;
337
        } else {
338
            h->top_cbp = 0;
339
        }
340
        // left_cbp
341
        if (left_type[0]) {
342
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
343
        } else if(IS_INTRA(mb_type)) {
344
            h->left_cbp = 0x1C0;
345
        } else {
346
            h->left_cbp = 0;
347
        }
348
        if (left_type[0]) {
349
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
350
        }
351
        if (left_type[1]) {
352
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
353
        }
354
    }
355

    
356
#if 1
357
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
358
        int list;
359
        for(list=0; list<h->list_count; list++){
360
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
361
                /*if(!h->mv_cache_clean[list]){
362
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
363
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
364
                    h->mv_cache_clean[list]= 1;
365
                }*/
366
                continue;
367
            }
368
            h->mv_cache_clean[list]= 0;
369

    
370
            if(USES_LIST(top_type, list)){
371
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
372
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
373
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
374
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
375
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
376
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
377
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
378
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
379
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
380
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
381
            }else{
382
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
383
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
384
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
385
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
386
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
387
            }
388

    
389
            for(i=0; i<2; i++){
390
                int cache_idx = scan8[0] - 1 + i*2*8;
391
                if(USES_LIST(left_type[i], list)){
392
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
393
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
394
                    *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
395
                    *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
396
                    h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
397
                    h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
398
                }else{
399
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
400
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
401
                    h->ref_cache[list][cache_idx  ]=
402
                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
403
                }
404
            }
405

    
406
            if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
407
                continue;
408

    
409
            if(USES_LIST(topleft_type, list)){
410
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
411
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
412
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
413
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
414
            }else{
415
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
416
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
417
            }
418

    
419
            if(USES_LIST(topright_type, list)){
420
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
421
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
422
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
423
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
424
            }else{
425
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
426
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
427
            }
428

    
429
            if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
430
                continue;
431

    
432
            h->ref_cache[list][scan8[5 ]+1] =
433
            h->ref_cache[list][scan8[7 ]+1] =
434
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
435
            h->ref_cache[list][scan8[4 ]] =
436
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
437
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
438
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
439
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
440
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
441
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
442

    
443
            if( h->pps.cabac ) {
444
                /* XXX beurk, Load mvd */
445
                if(USES_LIST(top_type, list)){
446
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
447
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
448
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
449
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
450
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
451
                }else{
452
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
453
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
454
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
455
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
456
                }
457
                if(USES_LIST(left_type[0], list)){
458
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
459
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
460
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
461
                }else{
462
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
463
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
464
                }
465
                if(USES_LIST(left_type[1], list)){
466
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
467
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
468
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
469
                }else{
470
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
471
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
472
                }
473
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
474
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
475
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
476
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
477
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
478

    
479
                if(h->slice_type_nos == FF_B_TYPE){
480
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
481

    
482
                    if(IS_DIRECT(top_type)){
483
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
484
                    }else if(IS_8X8(top_type)){
485
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
486
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
487
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
488
                    }else{
489
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
490
                    }
491

    
492
                    if(IS_DIRECT(left_type[0]))
493
                        h->direct_cache[scan8[0] - 1 + 0*8]= 1;
494
                    else if(IS_8X8(left_type[0]))
495
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
496
                    else
497
                        h->direct_cache[scan8[0] - 1 + 0*8]= 0;
498

    
499
                    if(IS_DIRECT(left_type[1]))
500
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
501
                    else if(IS_8X8(left_type[1]))
502
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
503
                    else
504
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
505
                }
506
            }
507

    
508
            if(FRAME_MBAFF){
509
#define MAP_MVS\
510
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
511
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
512
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
513
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
514
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
515
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
516
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
517
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
518
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
519
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
520
                if(MB_FIELD){
521
#define MAP_F2F(idx, mb_type)\
522
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
523
                        h->ref_cache[list][idx] <<= 1;\
524
                        h->mv_cache[list][idx][1] /= 2;\
525
                        h->mvd_cache[list][idx][1] /= 2;\
526
                    }
527
                    MAP_MVS
528
#undef MAP_F2F
529
                }else{
530
#define MAP_F2F(idx, mb_type)\
531
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
532
                        h->ref_cache[list][idx] >>= 1;\
533
                        h->mv_cache[list][idx][1] <<= 1;\
534
                        h->mvd_cache[list][idx][1] <<= 1;\
535
                    }
536
                    MAP_MVS
537
#undef MAP_F2F
538
                }
539
            }
540
        }
541
    }
542
#endif
543

    
544
    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
545
}
546

    
547
static inline void write_back_intra_pred_mode(H264Context *h){
548
    const int mb_xy= h->mb_xy;
549

    
550
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
551
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
552
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
553
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
554
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
555
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
556
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
557
}
558

    
559
/**
560
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
561
 */
562
static inline int check_intra4x4_pred_mode(H264Context *h){
563
    MpegEncContext * const s = &h->s;
564
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
565
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
566
    int i;
567

    
568
    if(!(h->top_samples_available&0x8000)){
569
        for(i=0; i<4; i++){
570
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
571
            if(status<0){
572
                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
573
                return -1;
574
            } else if(status){
575
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
576
            }
577
        }
578
    }
579

    
580
    if((h->left_samples_available&0x8888)!=0x8888){
581
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
582
        for(i=0; i<4; i++){
583
            if(!(h->left_samples_available&mask[i])){
584
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
585
                if(status<0){
586
                    av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
587
                    return -1;
588
                } else if(status){
589
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
590
                }
591
            }
592
        }
593
    }
594

    
595
    return 0;
596
} //FIXME cleanup like next
597

    
598
/**
599
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
600
 */
601
static inline int check_intra_pred_mode(H264Context *h, int mode){
602
    MpegEncContext * const s = &h->s;
603
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
604
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
605

    
606
    if(mode > 6U) {
607
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
608
        return -1;
609
    }
610

    
611
    if(!(h->top_samples_available&0x8000)){
612
        mode= top[ mode ];
613
        if(mode<0){
614
            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
615
            return -1;
616
        }
617
    }
618

    
619
    if((h->left_samples_available&0x8080) != 0x8080){
620
        mode= left[ mode ];
621
        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
622
            mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
623
        }
624
        if(mode<0){
625
            av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
626
            return -1;
627
        }
628
    }
629

    
630
    return mode;
631
}
632

    
633
/**
634
 * gets the predicted intra4x4 prediction mode.
635
 */
636
static inline int pred_intra_mode(H264Context *h, int n){
637
    const int index8= scan8[n];
638
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
639
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
640
    const int min= FFMIN(left, top);
641

    
642
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
643

    
644
    if(min<0) return DC_PRED;
645
    else      return min;
646
}
647

    
648
static inline void write_back_non_zero_count(H264Context *h){
649
    const int mb_xy= h->mb_xy;
650

    
651
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
652
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
653
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
654
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
655
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
656
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
657
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
658

    
659
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
660
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
661
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
662

    
663
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
664
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
665
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
666
}
667

    
668
/**
669
 * gets the predicted number of non-zero coefficients.
670
 * @param n block index
671
 */
672
static inline int pred_non_zero_count(H264Context *h, int n){
673
    const int index8= scan8[n];
674
    const int left= h->non_zero_count_cache[index8 - 1];
675
    const int top = h->non_zero_count_cache[index8 - 8];
676
    int i= left + top;
677

    
678
    if(i<64) i= (i+1)>>1;
679

    
680
    tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
681

    
682
    return i&31;
683
}
684

    
685
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
686
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
687
    MpegEncContext *s = &h->s;
688

    
689
    /* there is no consistent mapping of mvs to neighboring locations that will
690
     * make mbaff happy, so we can't move all this logic to fill_caches */
691
    if(FRAME_MBAFF){
692
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
693
        const int16_t *mv;
694
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
695
        *C = h->mv_cache[list][scan8[0]-2];
696

    
697
        if(!MB_FIELD
698
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
699
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
700
            if(IS_INTERLACED(mb_types[topright_xy])){
701
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
702
                const int x4 = X4, y4 = Y4;\
703
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
704
                if(!USES_LIST(mb_type,list))\
705
                    return LIST_NOT_USED;\
706
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
707
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
708
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
709
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
710

    
711
                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
712
            }
713
        }
714
        if(topright_ref == PART_NOT_AVAILABLE
715
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
716
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
717
            if(!MB_FIELD
718
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
719
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
720
            }
721
            if(MB_FIELD
722
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
723
               && i >= scan8[0]+8){
724
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
725
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
726
            }
727
        }
728
#undef SET_DIAG_MV
729
    }
730

    
731
    if(topright_ref != PART_NOT_AVAILABLE){
732
        *C= h->mv_cache[list][ i - 8 + part_width ];
733
        return topright_ref;
734
    }else{
735
        tprintf(s->avctx, "topright MV not available\n");
736

    
737
        *C= h->mv_cache[list][ i - 8 - 1 ];
738
        return h->ref_cache[list][ i - 8 - 1 ];
739
    }
740
}
741

    
742
/**
743
 * gets the predicted MV.
744
 * @param n the block index
745
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
746
 * @param mx the x component of the predicted motion vector
747
 * @param my the y component of the predicted motion vector
748
 */
749
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
750
    const int index8= scan8[n];
751
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
752
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
753
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
754
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
755
    const int16_t * C;
756
    int diagonal_ref, match_count;
757

    
758
    assert(part_width==1 || part_width==2 || part_width==4);
759

    
760
/* mv_cache
761
  B . . A T T T T
762
  U . . L . . , .
763
  U . . L . . . .
764
  U . . L . . , .
765
  . . . L . . . .
766
*/
767

    
768
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
769
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
770
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
771
    if(match_count > 1){ //most common
772
        *mx= mid_pred(A[0], B[0], C[0]);
773
        *my= mid_pred(A[1], B[1], C[1]);
774
    }else if(match_count==1){
775
        if(left_ref==ref){
776
            *mx= A[0];
777
            *my= A[1];
778
        }else if(top_ref==ref){
779
            *mx= B[0];
780
            *my= B[1];
781
        }else{
782
            *mx= C[0];
783
            *my= C[1];
784
        }
785
    }else{
786
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
787
            *mx= A[0];
788
            *my= A[1];
789
        }else{
790
            *mx= mid_pred(A[0], B[0], C[0]);
791
            *my= mid_pred(A[1], B[1], C[1]);
792
        }
793
    }
794

    
795
    tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
796
}
797

    
798
/**
799
 * gets the directionally predicted 16x8 MV.
800
 * @param n the block index
801
 * @param mx the x component of the predicted motion vector
802
 * @param my the y component of the predicted motion vector
803
 */
804
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
805
    if(n==0){
806
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
807
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
808

    
809
        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
810

    
811
        if(top_ref == ref){
812
            *mx= B[0];
813
            *my= B[1];
814
            return;
815
        }
816
    }else{
817
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
818
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
819

    
820
        tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
821

    
822
        if(left_ref == ref){
823
            *mx= A[0];
824
            *my= A[1];
825
            return;
826
        }
827
    }
828

    
829
    //RARE
830
    pred_motion(h, n, 4, list, ref, mx, my);
831
}
832

    
833
/**
834
 * gets the directionally predicted 8x16 MV.
835
 * @param n the block index
836
 * @param mx the x component of the predicted motion vector
837
 * @param my the y component of the predicted motion vector
838
 */
839
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
840
    if(n==0){
841
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
842
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
843

    
844
        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
845

    
846
        if(left_ref == ref){
847
            *mx= A[0];
848
            *my= A[1];
849
            return;
850
        }
851
    }else{
852
        const int16_t * C;
853
        int diagonal_ref;
854

    
855
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
856

    
857
        tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
858

    
859
        if(diagonal_ref == ref){
860
            *mx= C[0];
861
            *my= C[1];
862
            return;
863
        }
864
    }
865

    
866
    //RARE
867
    pred_motion(h, n, 2, list, ref, mx, my);
868
}
869

    
870
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
871
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
872
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
873

    
874
    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
875

    
876
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
877
       || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
878
       || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
879

    
880
        *mx = *my = 0;
881
        return;
882
    }
883

    
884
    pred_motion(h, 0, 4, 0, 0, mx, my);
885

    
886
    return;
887
}
888

    
889
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
890
    int poc0 = h->ref_list[0][i].poc;
891
    int td = av_clip(poc1 - poc0, -128, 127);
892
    if(td == 0 || h->ref_list[0][i].long_ref){
893
        return 256;
894
    }else{
895
        int tb = av_clip(poc - poc0, -128, 127);
896
        int tx = (16384 + (FFABS(td) >> 1)) / td;
897
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
898
    }
899
}
900

    
901
static inline void direct_dist_scale_factor(H264Context * const h){
902
    MpegEncContext * const s = &h->s;
903
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
904
    const int poc1 = h->ref_list[1][0].poc;
905
    int i, field;
906
    for(field=0; field<2; field++){
907
        const int poc  = h->s.current_picture_ptr->field_poc[field];
908
        const int poc1 = h->ref_list[1][0].field_poc[field];
909
        for(i=0; i < 2*h->ref_count[0]; i++)
910
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
911
    }
912

    
913
    for(i=0; i<h->ref_count[0]; i++){
914
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
915
    }
916
}
917

    
918
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
919
    MpegEncContext * const s = &h->s;
920
    Picture * const ref1 = &h->ref_list[1][0];
921
    int j, old_ref, rfield;
922
    int start= mbafi ? 16                      : 0;
923
    int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
924
    int interl= mbafi || s->picture_structure != PICT_FRAME;
925

    
926
    /* bogus; fills in for missing frames */
927
    memset(map[list], 0, sizeof(map[list]));
928

    
929
    for(rfield=0; rfield<2; rfield++){
930
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
931
            int poc = ref1->ref_poc[colfield][list][old_ref];
932

    
933
            if     (!interl)
934
                poc |= 3;
935
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
936
                poc= (poc&~3) + rfield + 1;
937

    
938
            for(j=start; j<end; j++){
939
                if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
940
                    int cur_ref= mbafi ? (j-16)^field : j;
941
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
942
                    if(rfield == field)
943
                        map[list][old_ref] = cur_ref;
944
                    break;
945
                }
946
            }
947
        }
948
    }
949
}
950

    
951
static inline void direct_ref_list_init(H264Context * const h){
952
    MpegEncContext * const s = &h->s;
953
    Picture * const ref1 = &h->ref_list[1][0];
954
    Picture * const cur = s->current_picture_ptr;
955
    int list, j, field;
956
    int sidx= (s->picture_structure&1)^1;
957
    int ref1sidx= (ref1->reference&1)^1;
958

    
959
    for(list=0; list<2; list++){
960
        cur->ref_count[sidx][list] = h->ref_count[list];
961
        for(j=0; j<h->ref_count[list]; j++)
962
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
963
    }
964

    
965
    if(s->picture_structure == PICT_FRAME){
966
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
967
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
968
    }
969

    
970
    cur->mbaff= FRAME_MBAFF;
971

    
972
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
973
        return;
974

    
975
    for(list=0; list<2; list++){
976
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
977
        for(field=0; field<2; field++)
978
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
979
    }
980
}
981

    
982
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
983
    MpegEncContext * const s = &h->s;
984
    int b8_stride = h->b8_stride;
985
    int b4_stride = h->b_stride;
986
    int mb_xy = h->mb_xy;
987
    int mb_type_col[2];
988
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
989
    const int8_t *l1ref0, *l1ref1;
990
    const int is_b8x8 = IS_8X8(*mb_type);
991
    unsigned int sub_mb_type;
992
    int i8, i4;
993

    
994
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
995

    
996
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
997
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
998
            int cur_poc = s->current_picture_ptr->poc;
999
            int *col_poc = h->ref_list[1]->field_poc;
1000
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1001
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1002
            b8_stride = 0;
1003
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1004
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1005
            mb_xy += s->mb_stride*fieldoff;
1006
        }
1007
        goto single_col;
1008
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
1009
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
1010
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1011
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1012
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1013
            b8_stride *= 3;
1014
            b4_stride *= 6;
1015
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1016
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1017
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1018
                && !is_b8x8){
1019
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1020
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1021
            }else{
1022
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1023
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1024
            }
1025
        }else{                                           //     AFR/FR    -> AFR/FR
1026
single_col:
1027
            mb_type_col[0] =
1028
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1029
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1030
                /* FIXME save sub mb types from previous frames (or derive from MVs)
1031
                * so we know exactly what block size to use */
1032
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1033
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1034
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1035
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1036
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1037
            }else{
1038
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1039
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
1040
            }
1041
        }
1042
    }
1043

    
1044
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1045
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1046
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1047
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1048
    if(!b8_stride){
1049
        if(s->mb_y&1){
1050
            l1ref0 += h->b8_stride;
1051
            l1ref1 += h->b8_stride;
1052
            l1mv0  +=  2*b4_stride;
1053
            l1mv1  +=  2*b4_stride;
1054
        }
1055
    }
1056

    
1057
    if(h->direct_spatial_mv_pred){
1058
        int ref[2];
1059
        int mv[2][2];
1060
        int list;
1061

    
1062
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1063

    
1064
        /* ref = min(neighbors) */
1065
        for(list=0; list<2; list++){
1066
            int refa = h->ref_cache[list][scan8[0] - 1];
1067
            int refb = h->ref_cache[list][scan8[0] - 8];
1068
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1069
            if(refc == PART_NOT_AVAILABLE)
1070
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1071
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1072
            if(ref[list] < 0)
1073
                ref[list] = -1;
1074
        }
1075

    
1076
        if(ref[0] < 0 && ref[1] < 0){
1077
            ref[0] = ref[1] = 0;
1078
            mv[0][0] = mv[0][1] =
1079
            mv[1][0] = mv[1][1] = 0;
1080
        }else{
1081
            for(list=0; list<2; list++){
1082
                if(ref[list] >= 0)
1083
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1084
                else
1085
                    mv[list][0] = mv[list][1] = 0;
1086
            }
1087
        }
1088

    
1089
        if(ref[1] < 0){
1090
            if(!is_b8x8)
1091
                *mb_type &= ~MB_TYPE_L1;
1092
            sub_mb_type &= ~MB_TYPE_L1;
1093
        }else if(ref[0] < 0){
1094
            if(!is_b8x8)
1095
                *mb_type &= ~MB_TYPE_L0;
1096
            sub_mb_type &= ~MB_TYPE_L0;
1097
        }
1098

    
1099
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1100
            for(i8=0; i8<4; i8++){
1101
                int x8 = i8&1;
1102
                int y8 = i8>>1;
1103
                int xy8 = x8+y8*b8_stride;
1104
                int xy4 = 3*x8+y8*b4_stride;
1105
                int a=0, b=0;
1106

    
1107
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1108
                    continue;
1109
                h->sub_mb_type[i8] = sub_mb_type;
1110

    
1111
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1112
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1113
                if(!IS_INTRA(mb_type_col[y8])
1114
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1115
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1116
                    if(ref[0] > 0)
1117
                        a= pack16to32(mv[0][0],mv[0][1]);
1118
                    if(ref[1] > 0)
1119
                        b= pack16to32(mv[1][0],mv[1][1]);
1120
                }else{
1121
                    a= pack16to32(mv[0][0],mv[0][1]);
1122
                    b= pack16to32(mv[1][0],mv[1][1]);
1123
                }
1124
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1125
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1126
            }
1127
        }else if(IS_16X16(*mb_type)){
1128
            int a=0, b=0;
1129

    
1130
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1131
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1132
            if(!IS_INTRA(mb_type_col[0])
1133
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1134
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1135
                       && (h->x264_build>33 || !h->x264_build)))){
1136
                if(ref[0] > 0)
1137
                    a= pack16to32(mv[0][0],mv[0][1]);
1138
                if(ref[1] > 0)
1139
                    b= pack16to32(mv[1][0],mv[1][1]);
1140
            }else{
1141
                a= pack16to32(mv[0][0],mv[0][1]);
1142
                b= pack16to32(mv[1][0],mv[1][1]);
1143
            }
1144
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1145
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1146
        }else{
1147
            for(i8=0; i8<4; i8++){
1148
                const int x8 = i8&1;
1149
                const int y8 = i8>>1;
1150

    
1151
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1152
                    continue;
1153
                h->sub_mb_type[i8] = sub_mb_type;
1154

    
1155
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1156
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1157
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1158
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1159

    
1160
                /* col_zero_flag */
1161
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
1162
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1163
                                                  && (h->x264_build>33 || !h->x264_build)))){
1164
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1165
                    if(IS_SUB_8X8(sub_mb_type)){
1166
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1167
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1168
                            if(ref[0] == 0)
1169
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1170
                            if(ref[1] == 0)
1171
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1172
                        }
1173
                    }else
1174
                    for(i4=0; i4<4; i4++){
1175
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1176
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1177
                            if(ref[0] == 0)
1178
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1179
                            if(ref[1] == 0)
1180
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1181
                        }
1182
                    }
1183
                }
1184
            }
1185
        }
1186
    }else{ /* direct temporal mv pred */
1187
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1188
        const int *dist_scale_factor = h->dist_scale_factor;
1189
        int ref_offset= 0;
1190

    
1191
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1192
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1193
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1194
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
1195
        }
1196
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1197
            ref_offset += 16;
1198

    
1199
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1200
            /* FIXME assumes direct_8x8_inference == 1 */
1201
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
1202

    
1203
            for(i8=0; i8<4; i8++){
1204
                const int x8 = i8&1;
1205
                const int y8 = i8>>1;
1206
                int ref0, scale;
1207
                const int16_t (*l1mv)[2]= l1mv0;
1208

    
1209
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1210
                    continue;
1211
                h->sub_mb_type[i8] = sub_mb_type;
1212

    
1213
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1214
                if(IS_INTRA(mb_type_col[y8])){
1215
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1216
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1217
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1218
                    continue;
1219
                }
1220

    
1221
                ref0 = l1ref0[x8 + y8*b8_stride];
1222
                if(ref0 >= 0)
1223
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
1224
                else{
1225
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1226
                    l1mv= l1mv1;
1227
                }
1228
                scale = dist_scale_factor[ref0];
1229
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1230

    
1231
                {
1232
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1233
                    int my_col = (mv_col[1]<<y_shift)/2;
1234
                    int mx = (scale * mv_col[0] + 128) >> 8;
1235
                    int my = (scale * my_col + 128) >> 8;
1236
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1237
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1238
                }
1239
            }
1240
            return;
1241
        }
1242

    
1243
        /* one-to-one mv scaling */
1244

    
1245
        if(IS_16X16(*mb_type)){
1246
            int ref, mv0, mv1;
1247

    
1248
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1249
            if(IS_INTRA(mb_type_col[0])){
1250
                ref=mv0=mv1=0;
1251
            }else{
1252
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1253
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
1254
                const int scale = dist_scale_factor[ref0];
1255
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1256
                int mv_l0[2];
1257
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1258
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1259
                ref= ref0;
1260
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
1261
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1262
            }
1263
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1264
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1265
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1266
        }else{
1267
            for(i8=0; i8<4; i8++){
1268
                const int x8 = i8&1;
1269
                const int y8 = i8>>1;
1270
                int ref0, scale;
1271
                const int16_t (*l1mv)[2]= l1mv0;
1272

    
1273
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1274
                    continue;
1275
                h->sub_mb_type[i8] = sub_mb_type;
1276
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1277
                if(IS_INTRA(mb_type_col[0])){
1278
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1279
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1280
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1281
                    continue;
1282
                }
1283

    
1284
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1285
                if(ref0 >= 0)
1286
                    ref0 = map_col_to_list0[0][ref0];
1287
                else{
1288
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1289
                    l1mv= l1mv1;
1290
                }
1291
                scale = dist_scale_factor[ref0];
1292

    
1293
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1294
                if(IS_SUB_8X8(sub_mb_type)){
1295
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1296
                    int mx = (scale * mv_col[0] + 128) >> 8;
1297
                    int my = (scale * mv_col[1] + 128) >> 8;
1298
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1299
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1300
                }else
1301
                for(i4=0; i4<4; i4++){
1302
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1303
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1304
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1305
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1306
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1307
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1308
                }
1309
            }
1310
        }
1311
    }
1312
}
1313

    
1314
static inline void write_back_motion(H264Context *h, int mb_type){
1315
    MpegEncContext * const s = &h->s;
1316
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1317
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1318
    int list;
1319

    
1320
    if(!USES_LIST(mb_type, 0))
1321
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1322

    
1323
    for(list=0; list<h->list_count; list++){
1324
        int y;
1325
        if(!USES_LIST(mb_type, list))
1326
            continue;
1327

    
1328
        for(y=0; y<4; y++){
1329
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1330
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1331
        }
1332
        if( h->pps.cabac ) {
1333
            if(IS_SKIP(mb_type))
1334
                fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1335
            else
1336
            for(y=0; y<4; y++){
1337
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1338
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1339
            }
1340
        }
1341

    
1342
        {
1343
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1344
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1345
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1346
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1347
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1348
        }
1349
    }
1350

    
1351
    if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1352
        if(IS_8X8(mb_type)){
1353
            uint8_t *direct_table = &h->direct_table[b8_xy];
1354
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1355
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1356
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1357
        }
1358
    }
1359
}
1360

    
1361
/**
1362
 * Decodes a network abstraction layer unit.
1363
 * @param consumed is the number of bytes used as input
1364
 * @param length is the length of the array
1365
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1366
 * @returns decoded bytes, might be src+1 if no escapes
1367
 */
1368
static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1369
    int i, si, di;
1370
    uint8_t *dst;
1371
    int bufidx;
1372

    
1373
//    src[0]&0x80;                //forbidden bit
1374
    h->nal_ref_idc= src[0]>>5;
1375
    h->nal_unit_type= src[0]&0x1F;
1376

    
1377
    src++; length--;
1378
#if 0
1379
    for(i=0; i<length; i++)
1380
        printf("%2X ", src[i]);
1381
#endif
1382

    
1383
#ifdef HAVE_FAST_UNALIGNED
1384
# ifdef HAVE_FAST_64BIT
1385
#   define RS 7
1386
    for(i=0; i+1<length; i+=9){
1387
        if(!((~*(uint64_t*)(src+i) & (*(uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1388
# else
1389
#   define RS 3
1390
    for(i=0; i+1<length; i+=5){
1391
        if(!((~*(uint32_t*)(src+i) & (*(uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1392
# endif
1393
            continue;
1394
        if(i>0 && !src[i]) i--;
1395
        while(src[i]) i++;
1396
#else
1397
#   define RS 0
1398
    for(i=0; i+1<length; i+=2){
1399
        if(src[i]) continue;
1400
        if(i>0 && src[i-1]==0) i--;
1401
#endif
1402
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1403
            if(src[i+2]!=3){
1404
                /* startcode, so we must be past the end */
1405
                length=i;
1406
            }
1407
            break;
1408
        }
1409
        i-= RS;
1410
    }
1411

    
1412
    if(i>=length-1){ //no escaped 0
1413
        *dst_length= length;
1414
        *consumed= length+1; //+1 for the header
1415
        return src;
1416
    }
1417

    
1418
    bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1419
    h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1420
    dst= h->rbsp_buffer[bufidx];
1421

    
1422
    if (dst == NULL){
1423
        return NULL;
1424
    }
1425

    
1426
//printf("decoding esc\n");
1427
    memcpy(dst, src, i);
1428
    si=di=i;
1429
    while(si+2<length){
1430
        //remove escapes (very rare 1:2^22)
1431
        if(src[si+2]>3){
1432
            dst[di++]= src[si++];
1433
            dst[di++]= src[si++];
1434
        }else if(src[si]==0 && src[si+1]==0){
1435
            if(src[si+2]==3){ //escape
1436
                dst[di++]= 0;
1437
                dst[di++]= 0;
1438
                si+=3;
1439
                continue;
1440
            }else //next start code
1441
                goto nsc;
1442
        }
1443

    
1444
        dst[di++]= src[si++];
1445
    }
1446
    while(si<length)
1447
        dst[di++]= src[si++];
1448
nsc:
1449

    
1450
    memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1451

    
1452
    *dst_length= di;
1453
    *consumed= si + 1;//+1 for the header
1454
//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1455
    return dst;
1456
}
1457

    
1458
/**
1459
 * identifies the exact end of the bitstream
1460
 * @return the length of the trailing, or 0 if damaged
1461
 */
1462
static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1463
    int v= *src;
1464
    int r;
1465

    
1466
    tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1467

    
1468
    for(r=1; r<9; r++){
1469
        if(v&1) return r;
1470
        v>>=1;
1471
    }
1472
    return 0;
1473
}
1474

    
1475
/**
1476
 * IDCT transforms the 16 dc values and dequantizes them.
1477
 * @param qp quantization parameter
1478
 */
1479
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1480
#define stride 16
1481
    int i;
1482
    int temp[16]; //FIXME check if this is a good idea
1483
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1484
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1485

    
1486
//memset(block, 64, 2*256);
1487
//return;
1488
    for(i=0; i<4; i++){
1489
        const int offset= y_offset[i];
1490
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1491
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1492
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1493
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1494

    
1495
        temp[4*i+0]= z0+z3;
1496
        temp[4*i+1]= z1+z2;
1497
        temp[4*i+2]= z1-z2;
1498
        temp[4*i+3]= z0-z3;
1499
    }
1500

    
1501
    for(i=0; i<4; i++){
1502
        const int offset= x_offset[i];
1503
        const int z0= temp[4*0+i] + temp[4*2+i];
1504
        const int z1= temp[4*0+i] - temp[4*2+i];
1505
        const int z2= temp[4*1+i] - temp[4*3+i];
1506
        const int z3= temp[4*1+i] + temp[4*3+i];
1507

    
1508
        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1509
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1510
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1511
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1512
    }
1513
}
1514

    
1515
#if 0
1516
/**
1517
 * DCT transforms the 16 dc values.
1518
 * @param qp quantization parameter ??? FIXME
1519
 */
1520
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1521
//    const int qmul= dequant_coeff[qp][0];
1522
    int i;
1523
    int temp[16]; //FIXME check if this is a good idea
1524
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1525
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1526

1527
    for(i=0; i<4; i++){
1528
        const int offset= y_offset[i];
1529
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1530
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1531
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1532
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1533

1534
        temp[4*i+0]= z0+z3;
1535
        temp[4*i+1]= z1+z2;
1536
        temp[4*i+2]= z1-z2;
1537
        temp[4*i+3]= z0-z3;
1538
    }
1539

1540
    for(i=0; i<4; i++){
1541
        const int offset= x_offset[i];
1542
        const int z0= temp[4*0+i] + temp[4*2+i];
1543
        const int z1= temp[4*0+i] - temp[4*2+i];
1544
        const int z2= temp[4*1+i] - temp[4*3+i];
1545
        const int z3= temp[4*1+i] + temp[4*3+i];
1546

1547
        block[stride*0 +offset]= (z0 + z3)>>1;
1548
        block[stride*2 +offset]= (z1 + z2)>>1;
1549
        block[stride*8 +offset]= (z1 - z2)>>1;
1550
        block[stride*10+offset]= (z0 - z3)>>1;
1551
    }
1552
}
1553
#endif
1554

    
1555
#undef xStride
1556
#undef stride
1557

    
1558
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1559
    const int stride= 16*2;
1560
    const int xStride= 16;
1561
    int a,b,c,d,e;
1562

    
1563
    a= block[stride*0 + xStride*0];
1564
    b= block[stride*0 + xStride*1];
1565
    c= block[stride*1 + xStride*0];
1566
    d= block[stride*1 + xStride*1];
1567

    
1568
    e= a-b;
1569
    a= a+b;
1570
    b= c-d;
1571
    c= c+d;
1572

    
1573
    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1574
    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1575
    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1576
    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1577
}
1578

    
1579
#if 0
1580
static void chroma_dc_dct_c(DCTELEM *block){
1581
    const int stride= 16*2;
1582
    const int xStride= 16;
1583
    int a,b,c,d,e;
1584

1585
    a= block[stride*0 + xStride*0];
1586
    b= block[stride*0 + xStride*1];
1587
    c= block[stride*1 + xStride*0];
1588
    d= block[stride*1 + xStride*1];
1589

1590
    e= a-b;
1591
    a= a+b;
1592
    b= c-d;
1593
    c= c+d;
1594

1595
    block[stride*0 + xStride*0]= (a+c);
1596
    block[stride*0 + xStride*1]= (e+b);
1597
    block[stride*1 + xStride*0]= (a-c);
1598
    block[stride*1 + xStride*1]= (e-b);
1599
}
1600
#endif
1601

    
1602
/**
1603
 * gets the chroma qp.
1604
 */
1605
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1606
    return h->pps.chroma_qp_table[t][qscale];
1607
}
1608

    
1609
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1610
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1611
                           int src_x_offset, int src_y_offset,
1612
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1613
    MpegEncContext * const s = &h->s;
1614
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1615
    int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1616
    const int luma_xy= (mx&3) + ((my&3)<<2);
1617
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1618
    uint8_t * src_cb, * src_cr;
1619
    int extra_width= h->emu_edge_width;
1620
    int extra_height= h->emu_edge_height;
1621
    int emu=0;
1622
    const int full_mx= mx>>2;
1623
    const int full_my= my>>2;
1624
    const int pic_width  = 16*s->mb_width;
1625
    const int pic_height = 16*s->mb_height >> MB_FIELD;
1626

    
1627
    if(mx&7) extra_width -= 3;
1628
    if(my&7) extra_height -= 3;
1629

    
1630
    if(   full_mx < 0-extra_width
1631
       || full_my < 0-extra_height
1632
       || full_mx + 16/*FIXME*/ > pic_width + extra_width
1633
       || full_my + 16/*FIXME*/ > pic_height + extra_height){
1634
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
1635
            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1636
        emu=1;
1637
    }
1638

    
1639
    qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1640
    if(!square){
1641
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1642
    }
1643

    
1644
    if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1645

    
1646
    if(MB_FIELD){
1647
        // chroma offset when predicting from a field of opposite parity
1648
        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1649
        emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1650
    }
1651
    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1652
    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1653

    
1654
    if(emu){
1655
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1656
            src_cb= s->edge_emu_buffer;
1657
    }
1658
    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1659

    
1660
    if(emu){
1661
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
1662
            src_cr= s->edge_emu_buffer;
1663
    }
1664
    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1665
}
1666

    
1667
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1668
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1669
                           int x_offset, int y_offset,
1670
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1671
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1672
                           int list0, int list1){
1673
    MpegEncContext * const s = &h->s;
1674
    qpel_mc_func *qpix_op=  qpix_put;
1675
    h264_chroma_mc_func chroma_op= chroma_put;
1676

    
1677
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1678
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1679
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1680
    x_offset += 8*s->mb_x;
1681
    y_offset += 8*(s->mb_y >> MB_FIELD);
1682

    
1683
    if(list0){
1684
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1685
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1686
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1687
                           qpix_op, chroma_op);
1688

    
1689
        qpix_op=  qpix_avg;
1690
        chroma_op= chroma_avg;
1691
    }
1692

    
1693
    if(list1){
1694
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1695
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1696
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1697
                           qpix_op, chroma_op);
1698
    }
1699
}
1700

    
1701
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1702
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1703
                           int x_offset, int y_offset,
1704
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1705
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1706
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1707
                           int list0, int list1){
1708
    MpegEncContext * const s = &h->s;
1709

    
1710
    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
1711
    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
1712
    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
1713
    x_offset += 8*s->mb_x;
1714
    y_offset += 8*(s->mb_y >> MB_FIELD);
1715

    
1716
    if(list0 && list1){
1717
        /* don't optimize for luma-only case, since B-frames usually
1718
         * use implicit weights => chroma too. */
1719
        uint8_t *tmp_cb = s->obmc_scratchpad;
1720
        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1721
        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1722
        int refn0 = h->ref_cache[0][ scan8[n] ];
1723
        int refn1 = h->ref_cache[1][ scan8[n] ];
1724

    
1725
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1726
                    dest_y, dest_cb, dest_cr,
1727
                    x_offset, y_offset, qpix_put, chroma_put);
1728
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1729
                    tmp_y, tmp_cb, tmp_cr,
1730
                    x_offset, y_offset, qpix_put, chroma_put);
1731

    
1732
        if(h->use_weight == 2){
1733
            int weight0 = h->implicit_weight[refn0][refn1];
1734
            int weight1 = 64 - weight0;
1735
            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
1736
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1737
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1738
        }else{
1739
            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1740
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1741
                            h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1742
            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1743
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1744
                            h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1745
            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1746
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1747
                            h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1748
        }
1749
    }else{
1750
        int list = list1 ? 1 : 0;
1751
        int refn = h->ref_cache[list][ scan8[n] ];
1752
        Picture *ref= &h->ref_list[list][refn];
1753
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1754
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
1755
                    qpix_put, chroma_put);
1756

    
1757
        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1758
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
1759
        if(h->use_weight_chroma){
1760
            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1761
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1762
            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1763
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1764
        }
1765
    }
1766
}
1767

    
1768
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1769
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1770
                           int x_offset, int y_offset,
1771
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1772
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1773
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1774
                           int list0, int list1){
1775
    if((h->use_weight==2 && list0 && list1
1776
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1777
       || h->use_weight==1)
1778
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1779
                         x_offset, y_offset, qpix_put, chroma_put,
1780
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1781
    else
1782
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1783
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1784
}
1785

    
1786
static inline void prefetch_motion(H264Context *h, int list){
1787
    /* fetch pixels for estimated mv 4 macroblocks ahead
1788
     * optimized for 64byte cache lines */
1789
    MpegEncContext * const s = &h->s;
1790
    const int refn = h->ref_cache[list][scan8[0]];
1791
    if(refn >= 0){
1792
        const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1793
        const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1794
        uint8_t **src= h->ref_list[list][refn].data;
1795
        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1796
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
1797
        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1798
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1799
    }
1800
}
1801

    
1802
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1803
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1804
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1805
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1806
    MpegEncContext * const s = &h->s;
1807
    const int mb_xy= h->mb_xy;
1808
    const int mb_type= s->current_picture.mb_type[mb_xy];
1809

    
1810
    assert(IS_INTER(mb_type));
1811

    
1812
    prefetch_motion(h, 0);
1813

    
1814
    if(IS_16X16(mb_type)){
1815
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1816
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1817
                &weight_op[0], &weight_avg[0],
1818
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1819
    }else if(IS_16X8(mb_type)){
1820
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1821
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1822
                &weight_op[1], &weight_avg[1],
1823
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1824
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1825
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1826
                &weight_op[1], &weight_avg[1],
1827
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1828
    }else if(IS_8X16(mb_type)){
1829
        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1830
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1831
                &weight_op[2], &weight_avg[2],
1832
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1833
        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1834
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1835
                &weight_op[2], &weight_avg[2],
1836
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1837
    }else{
1838
        int i;
1839

    
1840
        assert(IS_8X8(mb_type));
1841

    
1842
        for(i=0; i<4; i++){
1843
            const int sub_mb_type= h->sub_mb_type[i];
1844
            const int n= 4*i;
1845
            int x_offset= (i&1)<<2;
1846
            int y_offset= (i&2)<<1;
1847

    
1848
            if(IS_SUB_8X8(sub_mb_type)){
1849
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1850
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1851
                    &weight_op[3], &weight_avg[3],
1852
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1853
            }else if(IS_SUB_8X4(sub_mb_type)){
1854
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1855
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1856
                    &weight_op[4], &weight_avg[4],
1857
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1858
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1859
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1860
                    &weight_op[4], &weight_avg[4],
1861
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1862
            }else if(IS_SUB_4X8(sub_mb_type)){
1863
                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1864
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1865
                    &weight_op[5], &weight_avg[5],
1866
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1867
                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1868
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1869
                    &weight_op[5], &weight_avg[5],
1870
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1871
            }else{
1872
                int j;
1873
                assert(IS_SUB_4X4(sub_mb_type));
1874
                for(j=0; j<4; j++){
1875
                    int sub_x_offset= x_offset + 2*(j&1);
1876
                    int sub_y_offset= y_offset +   (j&2);
1877
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1878
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1879
                        &weight_op[6], &weight_avg[6],
1880
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1881
                }
1882
            }
1883
        }
1884
    }
1885

    
1886
    prefetch_motion(h, 1);
1887
}
1888

    
1889
static av_cold void init_cavlc_level_tab(void){
1890
    int suffix_length, mask;
1891
    unsigned int i;
1892

    
1893
    for(suffix_length=0; suffix_length<7; suffix_length++){
1894
        for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1895
            int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1896
            int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1897

    
1898
            mask= -(level_code&1);
1899
            level_code= (((2+level_code)>>1) ^ mask) - mask;
1900
            if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1901
                cavlc_level_tab[suffix_length][i][0]= level_code;
1902
                cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1903
            }else if(prefix + 1 <= LEVEL_TAB_BITS){
1904
                cavlc_level_tab[suffix_length][i][0]= prefix+100;
1905
                cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1906
            }else{
1907
                cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1908
                cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1909
            }
1910
        }
1911
    }
1912
}
1913

    
1914
static av_cold void decode_init_vlc(void){
1915
    static int done = 0;
1916

    
1917
    if (!done) {
1918
        int i;
1919
        int offset;
1920
        done = 1;
1921

    
1922
        chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1923
        chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1924
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1925
                 &chroma_dc_coeff_token_len [0], 1, 1,
1926
                 &chroma_dc_coeff_token_bits[0], 1, 1,
1927
                 INIT_VLC_USE_NEW_STATIC);
1928

    
1929
        offset = 0;
1930
        for(i=0; i<4; i++){
1931
            coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1932
            coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1933
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1934
                     &coeff_token_len [i][0], 1, 1,
1935
                     &coeff_token_bits[i][0], 1, 1,
1936
                     INIT_VLC_USE_NEW_STATIC);
1937
            offset += coeff_token_vlc_tables_size[i];
1938
        }
1939
        /*
1940
         * This is a one time safety check to make sure that
1941
         * the packed static coeff_token_vlc table sizes
1942
         * were initialized correctly.
1943
         */
1944
        assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1945

    
1946
        for(i=0; i<3; i++){
1947
            chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1948
            chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1949
            init_vlc(&chroma_dc_total_zeros_vlc[i],
1950
                     CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1951
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
1952
                     &chroma_dc_total_zeros_bits[i][0], 1, 1,
1953
                     INIT_VLC_USE_NEW_STATIC);
1954
        }
1955
        for(i=0; i<15; i++){
1956
            total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1957
            total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1958
            init_vlc(&total_zeros_vlc[i],
1959
                     TOTAL_ZEROS_VLC_BITS, 16,
1960
                     &total_zeros_len [i][0], 1, 1,
1961
                     &total_zeros_bits[i][0], 1, 1,
1962
                     INIT_VLC_USE_NEW_STATIC);
1963
        }
1964

    
1965
        for(i=0; i<6; i++){
1966
            run_vlc[i].table = run_vlc_tables[i];
1967
            run_vlc[i].table_allocated = run_vlc_tables_size;
1968
            init_vlc(&run_vlc[i],
1969
                     RUN_VLC_BITS, 7,
1970
                     &run_len [i][0], 1, 1,
1971
                     &run_bits[i][0], 1, 1,
1972
                     INIT_VLC_USE_NEW_STATIC);
1973
        }
1974
        run7_vlc.table = run7_vlc_table,
1975
        run7_vlc.table_allocated = run7_vlc_table_size;
1976
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1977
                 &run_len [6][0], 1, 1,
1978
                 &run_bits[6][0], 1, 1,
1979
                 INIT_VLC_USE_NEW_STATIC);
1980

    
1981
        init_cavlc_level_tab();
1982
    }
1983
}
1984

    
1985
static void free_tables(H264Context *h){
1986
    int i;
1987
    H264Context *hx;
1988
    av_freep(&h->intra4x4_pred_mode);
1989
    av_freep(&h->chroma_pred_mode_table);
1990
    av_freep(&h->cbp_table);
1991
    av_freep(&h->mvd_table[0]);
1992
    av_freep(&h->mvd_table[1]);
1993
    av_freep(&h->direct_table);
1994
    av_freep(&h->non_zero_count);
1995
    av_freep(&h->slice_table_base);
1996
    h->slice_table= NULL;
1997

    
1998
    av_freep(&h->mb2b_xy);
1999
    av_freep(&h->mb2b8_xy);
2000

    
2001
    for(i = 0; i < h->s.avctx->thread_count; i++) {
2002
        hx = h->thread_context[i];
2003
        if(!hx) continue;
2004
        av_freep(&hx->top_borders[1]);
2005
        av_freep(&hx->top_borders[0]);
2006
        av_freep(&hx->s.obmc_scratchpad);
2007
    }
2008
}
2009

    
2010
static void init_dequant8_coeff_table(H264Context *h){
2011
    int i,q,x;
2012
    const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2013
    h->dequant8_coeff[0] = h->dequant8_buffer[0];
2014
    h->dequant8_coeff[1] = h->dequant8_buffer[1];
2015

    
2016
    for(i=0; i<2; i++ ){
2017
        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2018
            h->dequant8_coeff[1] = h->dequant8_buffer[0];
2019
            break;
2020
        }
2021

    
2022
        for(q=0; q<52; q++){
2023
            int shift = div6[q];
2024
            int idx = rem6[q];
2025
            for(x=0; x<64; x++)
2026
                h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2027
                    ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2028
                    h->pps.scaling_matrix8[i][x]) << shift;
2029
        }
2030
    }
2031
}
2032

    
2033
static void init_dequant4_coeff_table(H264Context *h){
2034
    int i,j,q,x;
2035
    const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2036
    for(i=0; i<6; i++ ){
2037
        h->dequant4_coeff[i] = h->dequant4_buffer[i];
2038
        for(j=0; j<i; j++){
2039
            if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2040
                h->dequant4_coeff[i] = h->dequant4_buffer[j];
2041
                break;
2042
            }
2043
        }
2044
        if(j<i)
2045
            continue;
2046

    
2047
        for(q=0; q<52; q++){
2048
            int shift = div6[q] + 2;
2049
            int idx = rem6[q];
2050
            for(x=0; x<16; x++)
2051
                h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2052
                    ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2053
                    h->pps.scaling_matrix4[i][x]) << shift;
2054
        }
2055
    }
2056
}
2057

    
2058
static void init_dequant_tables(H264Context *h){
2059
    int i,x;
2060
    init_dequant4_coeff_table(h);
2061
    if(h->pps.transform_8x8_mode)
2062
        init_dequant8_coeff_table(h);
2063
    if(h->sps.transform_bypass){
2064
        for(i=0; i<6; i++)
2065
            for(x=0; x<16; x++)
2066
                h->dequant4_coeff[i][0][x] = 1<<6;
2067
        if(h->pps.transform_8x8_mode)
2068
            for(i=0; i<2; i++)
2069
                for(x=0; x<64; x++)
2070
                    h->dequant8_coeff[i][0][x] = 1<<6;
2071
    }
2072
}
2073

    
2074

    
2075
/**
2076
 * allocates tables.
2077
 * needs width/height
2078
 */
2079
static int alloc_tables(H264Context *h){
2080
    MpegEncContext * const s = &h->s;
2081
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2082
    int x,y;
2083

    
2084
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2085

    
2086
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2087
    CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2088
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2089

    
2090
    CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2091
    CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2092
    CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2093
    CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2094

    
2095
    memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(*h->slice_table_base));
2096
    h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2097

    
2098
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));
2099
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2100
    for(y=0; y<s->mb_height; y++){
2101
        for(x=0; x<s->mb_width; x++){
2102
            const int mb_xy= x + y*s->mb_stride;
2103
            const int b_xy = 4*x + 4*y*h->b_stride;
2104
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2105

    
2106
            h->mb2b_xy [mb_xy]= b_xy;
2107
            h->mb2b8_xy[mb_xy]= b8_xy;
2108
        }
2109
    }
2110

    
2111
    s->obmc_scratchpad = NULL;
2112

    
2113
    if(!h->dequant4_coeff[0])
2114
        init_dequant_tables(h);
2115

    
2116
    return 0;
2117
fail:
2118
    free_tables(h);
2119
    return -1;
2120
}
2121

    
2122
/**
2123
 * Mimic alloc_tables(), but for every context thread.
2124
 */
2125
static void clone_tables(H264Context *dst, H264Context *src){
2126
    dst->intra4x4_pred_mode       = src->intra4x4_pred_mode;
2127
    dst->non_zero_count           = src->non_zero_count;
2128
    dst->slice_table              = src->slice_table;
2129
    dst->cbp_table                = src->cbp_table;
2130
    dst->mb2b_xy                  = src->mb2b_xy;
2131
    dst->mb2b8_xy                 = src->mb2b8_xy;
2132
    dst->chroma_pred_mode_table   = src->chroma_pred_mode_table;
2133
    dst->mvd_table[0]             = src->mvd_table[0];
2134
    dst->mvd_table[1]             = src->mvd_table[1];
2135
    dst->direct_table             = src->direct_table;
2136

    
2137
    dst->s.obmc_scratchpad = NULL;
2138
    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2139
}
2140

    
2141
/**
2142
 * Init context
2143
 * Allocate buffers which are not shared amongst multiple threads.
2144
 */
2145
static int context_init(H264Context *h){
2146
    CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2147
    CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2148

    
2149
    return 0;
2150
fail:
2151
    return -1; // free_tables will clean up for us
2152
}
2153

    
2154
static av_cold void common_init(H264Context *h){
2155
    MpegEncContext * const s = &h->s;
2156

    
2157
    s->width = s->avctx->width;
2158
    s->height = s->avctx->height;
2159
    s->codec_id= s->avctx->codec->id;
2160

    
2161
    ff_h264_pred_init(&h->hpc, s->codec_id);
2162

    
2163
    h->dequant_coeff_pps= -1;
2164
    s->unrestricted_mv=1;
2165
    s->decode=1; //FIXME
2166

    
2167
    dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2168

    
2169
    memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2170
    memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2171
}
2172

    
2173
static av_cold int decode_init(AVCodecContext *avctx){
2174
    H264Context *h= avctx->priv_data;
2175
    MpegEncContext * const s = &h->s;
2176

    
2177
    MPV_decode_defaults(s);
2178

    
2179
    s->avctx = avctx;
2180
    common_init(h);
2181

    
2182
    s->out_format = FMT_H264;
2183
    s->workaround_bugs= avctx->workaround_bugs;
2184

    
2185
    // set defaults
2186
//    s->decode_mb= ff_h263_decode_mb;
2187
    s->quarter_sample = 1;
2188
    s->low_delay= 1;
2189

    
2190
    if(avctx->codec_id == CODEC_ID_SVQ3)
2191
        avctx->pix_fmt= PIX_FMT_YUVJ420P;
2192
    else if(avctx->codec_id == CODEC_ID_H264_VDPAU)
2193
        avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2194
    else
2195
        avctx->pix_fmt= PIX_FMT_YUV420P;
2196

    
2197
    decode_init_vlc();
2198

    
2199
    if(avctx->extradata_size > 0 && avctx->extradata &&
2200
       *(char *)avctx->extradata == 1){
2201
        h->is_avc = 1;
2202
        h->got_avcC = 0;
2203
    } else {
2204
        h->is_avc = 0;
2205
    }
2206

    
2207
    h->thread_context[0] = h;
2208
    h->outputed_poc = INT_MIN;
2209
    h->prev_poc_msb= 1<<16;
2210
    return 0;
2211
}
2212

    
2213
static int frame_start(H264Context *h){
2214
    MpegEncContext * const s = &h->s;
2215
    int i;
2216

    
2217
    if(MPV_frame_start(s, s->avctx) < 0)
2218
        return -1;
2219
    ff_er_frame_start(s);
2220
    /*
2221
     * MPV_frame_start uses pict_type to derive key_frame.
2222
     * This is incorrect for H.264; IDR markings must be used.
2223
     * Zero here; IDR markings per slice in frame or fields are ORed in later.
2224
     * See decode_nal_units().
2225
     */
2226
    s->current_picture_ptr->key_frame= 0;
2227

    
2228
    assert(s->linesize && s->uvlinesize);
2229

    
2230
    for(i=0; i<16; i++){
2231
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2232
        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2233
    }
2234
    for(i=0; i<4; i++){
2235
        h->block_offset[16+i]=
2236
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2237
        h->block_offset[24+16+i]=
2238
        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2239
    }
2240

    
2241
    /* can't be in alloc_tables because linesize isn't known there.
2242
     * FIXME: redo bipred weight to not require extra buffer? */
2243
    for(i = 0; i < s->avctx->thread_count; i++)
2244
        if(!h->thread_context[i]->s.obmc_scratchpad)
2245
            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2246

    
2247
    /* some macroblocks will be accessed before they're available */
2248
    if(FRAME_MBAFF || s->avctx->thread_count > 1)
2249
        memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2250

    
2251
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2252

    
2253
    // We mark the current picture as non-reference after allocating it, so
2254
    // that if we break out due to an error it can be released automatically
2255
    // in the next MPV_frame_start().
2256
    // SVQ3 as well as most other codecs have only last/next/current and thus
2257
    // get released even with set reference, besides SVQ3 and others do not
2258
    // mark frames as reference later "naturally".
2259
    if(s->codec_id != CODEC_ID_SVQ3)
2260
        s->current_picture_ptr->reference= 0;
2261

    
2262
    s->current_picture_ptr->field_poc[0]=
2263
    s->current_picture_ptr->field_poc[1]= INT_MAX;
2264
    assert(s->current_picture_ptr->long_ref==0);
2265

    
2266
    return 0;
2267
}
2268

    
2269
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
2270
    MpegEncContext * const s = &h->s;
2271
    int i;
2272
    int step    = 1;
2273
    int offset  = 1;
2274
    int uvoffset= 1;
2275
    int top_idx = 1;
2276
    int skiplast= 0;
2277

    
2278
    src_y  -=   linesize;
2279
    src_cb -= uvlinesize;
2280
    src_cr -= uvlinesize;
2281

    
2282
    if(!simple && FRAME_MBAFF){
2283
        if(s->mb_y&1){
2284
            offset  = MB_MBAFF ? 1 : 17;
2285
            uvoffset= MB_MBAFF ? 1 : 9;
2286
            if(!MB_MBAFF){
2287
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y +  15*linesize);
2288
                *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2289
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2290
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2291
                    *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2292
                }
2293
            }
2294
        }else{
2295
            if(!MB_MBAFF){
2296
                h->left_border[0]= h->top_borders[0][s->mb_x][15];
2297
                if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2298
                    h->left_border[34   ]= h->top_borders[0][s->mb_x][16+7  ];
2299
                    h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2300
                }
2301
                skiplast= 1;
2302
            }
2303
            offset  =
2304
            uvoffset=
2305
            top_idx = MB_MBAFF ? 0 : 1;
2306
        }
2307
        step= MB_MBAFF ? 2 : 1;
2308
    }
2309

    
2310
    // There are two lines saved, the line above the the top macroblock of a pair,
2311
    // and the line above the bottom macroblock
2312
    h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2313
    for(i=1; i<17 - skiplast; i++){
2314
        h->left_border[offset+i*step]= src_y[15+i*  linesize];
2315
    }
2316

    
2317
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2318
    *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2319

    
2320
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2321
        h->left_border[uvoffset+34   ]= h->top_borders[top_idx][s->mb_x][16+7];
2322
        h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2323
        for(i=1; i<9 - skiplast; i++){
2324
            h->left_border[uvoffset+34   +i*step]= src_cb[7+i*uvlinesize];
2325
            h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2326
        }
2327
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2328
        *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2329
    }
2330
}
2331

    
2332
static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
2333
    MpegEncContext * const s = &h->s;
2334
    int temp8, i;
2335
    uint64_t temp64;
2336
    int deblock_left;
2337
    int deblock_top;
2338
    int mb_xy;
2339
    int step    = 1;
2340
    int offset  = 1;
2341
    int uvoffset= 1;
2342
    int top_idx = 1;
2343

    
2344
    if(!simple && FRAME_MBAFF){
2345
        if(s->mb_y&1){
2346
            offset  = MB_MBAFF ? 1 : 17;
2347
            uvoffset= MB_MBAFF ? 1 : 9;
2348
        }else{
2349
            offset  =
2350
            uvoffset=
2351
            top_idx = MB_MBAFF ? 0 : 1;
2352
        }
2353
        step= MB_MBAFF ? 2 : 1;
2354
    }
2355

    
2356
    if(h->deblocking_filter == 2) {
2357
        mb_xy = h->mb_xy;
2358
        deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2359
        deblock_top  = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2360
    } else {
2361
        deblock_left = (s->mb_x > 0);
2362
        deblock_top =  (s->mb_y > !!MB_FIELD);
2363
    }
2364

    
2365
    src_y  -=   linesize + 1;
2366
    src_cb -= uvlinesize + 1;
2367
    src_cr -= uvlinesize + 1;
2368

    
2369
#define XCHG(a,b,t,xchg)\
2370
t= a;\
2371
if(xchg)\
2372
    a= b;\
2373
b= t;
2374

    
2375
    if(deblock_left){
2376
        for(i = !deblock_top; i<16; i++){
2377
            XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, xchg);
2378
        }
2379
        XCHG(h->left_border[offset+i*step], src_y [i*  linesize], temp8, 1);
2380
    }
2381

    
2382
    if(deblock_top){
2383
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2384
        XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2385
        if(s->mb_x+1 < s->mb_width){
2386
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2387
        }
2388
    }
2389

    
2390
    if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2391
        if(deblock_left){
2392
            for(i = !deblock_top; i<8; i++){
2393
                XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, xchg);
2394
                XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2395
            }
2396
            XCHG(h->left_border[uvoffset+34   +i*step], src_cb[i*uvlinesize], temp8, 1);
2397
            XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2398
        }
2399
        if(deblock_top){
2400
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2401
            XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2402
        }
2403
    }
2404
}
2405

    
2406
static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2407
    MpegEncContext * const s = &h->s;
2408
    const int mb_x= s->mb_x;
2409
    const int mb_y= s->mb_y;
2410
    const int mb_xy= h->mb_xy;
2411
    const int mb_type= s->current_picture.mb_type[mb_xy];
2412
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2413
    int linesize, uvlinesize /*dct_offset*/;
2414
    int i;
2415
    int *block_offset = &h->block_offset[0];
2416
    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2417
    /* is_h264 should always be true if SVQ3 is disabled. */
2418
    const int is_h264 = !ENABLE_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2419
    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2420
    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2421

    
2422
    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
2423
    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2424
    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2425

    
2426
    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2427
    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2428

    
2429
    if (!simple && MB_FIELD) {
2430
        linesize   = h->mb_linesize   = s->linesize * 2;
2431
        uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2432
        block_offset = &h->block_offset[24];
2433
        if(mb_y&1){ //FIXME move out of this function?
2434
            dest_y -= s->linesize*15;
2435
            dest_cb-= s->uvlinesize*7;
2436
            dest_cr-= s->uvlinesize*7;
2437
        }
2438
        if(FRAME_MBAFF) {
2439
            int list;
2440
            for(list=0; list<h->list_count; list++){
2441
                if(!USES_LIST(mb_type, list))
2442
                    continue;
2443
                if(IS_16X16(mb_type)){
2444
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
2445
                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2446
                }else{
2447
                    for(i=0; i<16; i+=4){
2448
                        int ref = h->ref_cache[list][scan8[i]];
2449
                        if(ref >= 0)
2450
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2451
                    }
2452
                }
2453
            }
2454
        }
2455
    } else {
2456
        linesize   = h->mb_linesize   = s->linesize;
2457
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2458
//        dct_offset = s->linesize * 16;
2459
    }
2460

    
2461
    if (!simple && IS_INTRA_PCM(mb_type)) {
2462
        for (i=0; i<16; i++) {
2463
            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
2464
        }
2465
        for (i=0; i<8; i++) {
2466
            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
2467
            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
2468
        }
2469
    } else {
2470
        if(IS_INTRA(mb_type)){
2471
            if(h->deblocking_filter)
2472
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2473

    
2474
            if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2475
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2476
                h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2477
            }
2478

    
2479
            if(IS_INTRA4x4(mb_type)){
2480
                if(simple || !s->encoding){
2481
                    if(IS_8x8DCT(mb_type)){
2482
                        if(transform_bypass){
2483
                            idct_dc_add =
2484
                            idct_add    = s->dsp.add_pixels8;
2485
                        }else{
2486
                            idct_dc_add = s->dsp.h264_idct8_dc_add;
2487
                            idct_add    = s->dsp.h264_idct8_add;
2488
                        }
2489
                        for(i=0; i<16; i+=4){
2490
                            uint8_t * const ptr= dest_y + block_offset[i];
2491
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2492
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2493
                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2494
                            }else{
2495
                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
2496
                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2497
                                                            (h->topright_samples_available<<i)&0x4000, linesize);
2498
                                if(nnz){
2499
                                    if(nnz == 1 && h->mb[i*16])
2500
                                        idct_dc_add(ptr, h->mb + i*16, linesize);
2501
                                    else
2502
                                        idct_add   (ptr, h->mb + i*16, linesize);
2503
                                }
2504
                            }
2505
                        }
2506
                    }else{
2507
                        if(transform_bypass){
2508
                            idct_dc_add =
2509
                            idct_add    = s->dsp.add_pixels4;
2510
                        }else{
2511
                            idct_dc_add = s->dsp.h264_idct_dc_add;
2512
                            idct_add    = s->dsp.h264_idct_add;
2513
                        }
2514
                        for(i=0; i<16; i++){
2515
                            uint8_t * const ptr= dest_y + block_offset[i];
2516
                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2517

    
2518
                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2519
                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2520
                            }else{
2521
                                uint8_t *topright;
2522
                                int nnz, tr;
2523
                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2524
                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2525
                                    assert(mb_y || linesize <= block_offset[i]);
2526
                                    if(!topright_avail){
2527
                                        tr= ptr[3 - linesize]*0x01010101;
2528
                                        topright= (uint8_t*) &tr;
2529
                                    }else
2530
                                        topright= ptr + 4 - linesize;
2531
                                }else
2532
                                    topright= NULL;
2533

    
2534
                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2535
                                nnz = h->non_zero_count_cache[ scan8[i] ];
2536
                                if(nnz){
2537
                                    if(is_h264){
2538
                                        if(nnz == 1 && h->mb[i*16])
2539
                                            idct_dc_add(ptr, h->mb + i*16, linesize);
2540
                                        else
2541
                                            idct_add   (ptr, h->mb + i*16, linesize);
2542
                                    }else
2543
                                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2544
                                }
2545
                            }
2546
                        }
2547
                    }
2548
                }
2549
            }else{
2550
                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2551
                if(is_h264){
2552
                    if(!transform_bypass)
2553
                        h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
2554
                }else
2555
                    svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2556
            }
2557
            if(h->deblocking_filter)
2558
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
2559
        }else if(is_h264){
2560
            hl_motion(h, dest_y, dest_cb, dest_cr,
2561
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2562
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2563
                      s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2564
        }
2565

    
2566

    
2567
        if(!IS_INTRA4x4(mb_type)){
2568
            if(is_h264){
2569
                if(IS_INTRA16x16(mb_type)){
2570
                    if(transform_bypass){
2571
                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
2572
                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
2573
                        }else{
2574
                            for(i=0; i<16; i++){
2575
                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2576
                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
2577
                            }
2578
                        }
2579
                    }else{
2580
                         s->dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2581
                    }
2582
                }else if(h->cbp&15){
2583
                    if(transform_bypass){
2584
                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2585
                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2586
                        for(i=0; i<16; i+=di){
2587
                            if(h->non_zero_count_cache[ scan8[i] ]){
2588
                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2589
                            }
2590
                        }
2591
                    }else{
2592
                        if(IS_8x8DCT(mb_type)){
2593
                            s->dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2594
                        }else{
2595
                            s->dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
2596
                        }
2597
                    }
2598
                }
2599
            }else{
2600
                for(i=0; i<16; i++){
2601
                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2602
                        uint8_t * const ptr= dest_y + block_offset[i];
2603
                        svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2604
                    }
2605
                }
2606
            }
2607
        }
2608

    
2609
        if((simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2610
            uint8_t *dest[2] = {dest_cb, dest_cr};
2611
            if(transform_bypass){
2612
                if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2613
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
2614
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
2615
                }else{
2616
                    idct_add = s->dsp.add_pixels4;
2617
                    for(i=16; i<16+8; i++){
2618
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
2619
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2620
                    }
2621
                }
2622
            }else{
2623
                chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2624
                chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2625
                if(is_h264){
2626
                    idct_add = s->dsp.h264_idct_add;
2627
                    idct_dc_add = s->dsp.h264_idct_dc_add;
2628
                    for(i=16; i<16+8; i++){
2629
                        if(h->non_zero_count_cache[ scan8[i] ])
2630
                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2631
                        else if(h->mb[i*16])
2632
                            idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2633
                    }
2634
                }else{
2635
                    for(i=16; i<16+8; i++){
2636
                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2637
                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
2638
                            svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2639
                        }
2640
                    }
2641
                }
2642
            }
2643
        }
2644
    }
2645
    if(h->cbp || IS_INTRA(mb_type))
2646
        s->dsp.clear_blocks(h->mb);
2647

    
2648
    if(h->deblocking_filter) {
2649
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
2650
        fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
2651
        h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2652
        h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
2653
        if (!simple && FRAME_MBAFF) {
2654
            filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2655
        } else {
2656
            filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2657
        }
2658
    }
2659
}
2660

    
2661
/**
2662
 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2663
 */
2664
static void hl_decode_mb_simple(H264Context *h){
2665
    hl_decode_mb_internal(h, 1);
2666
}
2667

    
2668
/**
2669
 * Process a macroblock; this handles edge cases, such as interlacing.
2670
 */
2671
static void av_noinline hl_decode_mb_complex(H264Context *h){
2672
    hl_decode_mb_internal(h, 0);
2673
}
2674

    
2675
static void hl_decode_mb(H264Context *h){
2676
    MpegEncContext * const s = &h->s;
2677
    const int mb_xy= h->mb_xy;
2678
    const int mb_type= s->current_picture.mb_type[mb_xy];
2679
    int is_complex = ENABLE_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2680

    
2681
    if(ENABLE_H264_ENCODER && !s->decode)
2682
        return;
2683

    
2684
    if (is_complex)
2685
        hl_decode_mb_complex(h);
2686
    else hl_decode_mb_simple(h);
2687
}
2688

    
2689
static void pic_as_field(Picture *pic, const int parity){
2690
    int i;
2691
    for (i = 0; i < 4; ++i) {
2692
        if (parity == PICT_BOTTOM_FIELD)
2693
            pic->data[i] += pic->linesize[i];
2694
        pic->reference = parity;
2695
        pic->linesize[i] *= 2;
2696
    }
2697
    pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
2698
}
2699

    
2700
static int split_field_copy(Picture *dest, Picture *src,
2701
                            int parity, int id_add){
2702
    int match = !!(src->reference & parity);
2703

    
2704
    if (match) {
2705
        *dest = *src;
2706
        if(parity != PICT_FRAME){
2707
            pic_as_field(dest, parity);
2708
            dest->pic_id *= 2;
2709
            dest->pic_id += id_add;
2710
        }
2711
    }
2712

    
2713
    return match;
2714
}
2715

    
2716
static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2717
    int i[2]={0};
2718
    int index=0;
2719

    
2720
    while(i[0]<len || i[1]<len){
2721
        while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2722
            i[0]++;
2723
        while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2724
            i[1]++;
2725
        if(i[0] < len){
2726
            in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2727
            split_field_copy(&def[index++], in[ i[0]++ ], sel  , 1);
2728
        }
2729
        if(i[1] < len){
2730
            in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2731
            split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
2732
        }
2733
    }
2734

    
2735
    return index;
2736
}
2737

    
2738
static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2739
    int i, best_poc;
2740
    int out_i= 0;
2741

    
2742
    for(;;){
2743
        best_poc= dir ? INT_MIN : INT_MAX;
2744

    
2745
        for(i=0; i<len; i++){
2746
            const int poc= src[i]->poc;
2747
            if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2748
                best_poc= poc;
2749
                sorted[out_i]= src[i];
2750
            }
2751
        }
2752
        if(best_poc == (dir ? INT_MIN : INT_MAX))
2753
            break;
2754
        limit= sorted[out_i++]->poc - dir;
2755
    }
2756
    return out_i;
2757
}
2758

    
2759
/**
2760
 * fills the default_ref_list.
2761
 */
2762
static int fill_default_ref_list(H264Context *h){
2763
    MpegEncContext * const s = &h->s;
2764
    int i, len;
2765

    
2766
    if(h->slice_type_nos==FF_B_TYPE){
2767
        Picture *sorted[32];
2768
        int cur_poc, list;
2769
        int lens[2];
2770

    
2771
        if(FIELD_PICTURE)
2772
            cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2773
        else
2774
            cur_poc= s->current_picture_ptr->poc;
2775

    
2776
        for(list= 0; list<2; list++){
2777
            len= add_sorted(sorted    , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2778
            len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2779
            assert(len<=32);
2780
            len= build_def_list(h->default_ref_list[list]    , sorted     , len, 0, s->picture_structure);
2781
            len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2782
            assert(len<=32);
2783

    
2784
            if(len < h->ref_count[list])
2785
                memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2786
            lens[list]= len;
2787
        }
2788

    
2789
        if(lens[0] == lens[1] && lens[1] > 1){
2790
            for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2791
            if(i == lens[0])
2792
                FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
2793
        }
2794
    }else{
2795
        len = build_def_list(h->default_ref_list[0]    , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2796
        len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16                , 1, s->picture_structure);
2797
        assert(len <= 32);
2798
        if(len < h->ref_count[0])
2799
            memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
2800
    }
2801
#ifdef TRACE
2802
    for (i=0; i<h->ref_count[0]; i++) {
2803
        tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2804
    }
2805
    if(h->slice_type_nos==FF_B_TYPE){
2806
        for (i=0; i<h->ref_count[1]; i++) {
2807
            tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[1][i].data[0]);
2808
        }
2809
    }
2810
#endif
2811
    return 0;
2812
}
2813

    
2814
static void print_short_term(H264Context *h);
2815
static void print_long_term(H264Context *h);
2816

    
2817
/**
2818
 * Extract structure information about the picture described by pic_num in
2819
 * the current decoding context (frame or field). Note that pic_num is
2820
 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2821
 * @param pic_num picture number for which to extract structure information
2822
 * @param structure one of PICT_XXX describing structure of picture
2823
 *                      with pic_num
2824
 * @return frame number (short term) or long term index of picture
2825
 *         described by pic_num
2826
 */
2827
static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2828
    MpegEncContext * const s = &h->s;
2829

    
2830
    *structure = s->picture_structure;
2831
    if(FIELD_PICTURE){
2832
        if (!(pic_num & 1))
2833
            /* opposite field */
2834
            *structure ^= PICT_FRAME;
2835
        pic_num >>= 1;
2836
    }
2837

    
2838
    return pic_num;
2839
}
2840

    
2841
static int decode_ref_pic_list_reordering(H264Context *h){
2842
    MpegEncContext * const s = &h->s;
2843
    int list, index, pic_structure;
2844

    
2845
    print_short_term(h);
2846
    print_long_term(h);
2847

    
2848
    for(list=0; list<h->list_count; list++){
2849
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2850

    
2851
        if(get_bits1(&s->gb)){
2852
            int pred= h->curr_pic_num;
2853

    
2854
            for(index=0; ; index++){
2855
                unsigned int reordering_of_pic_nums_idc= get_ue_golomb_31(&s->gb);
2856
                unsigned int pic_id;
2857
                int i;
2858
                Picture *ref = NULL;
2859

    
2860
                if(reordering_of_pic_nums_idc==3)
2861
                    break;
2862

    
2863
                if(index >= h->ref_count[list]){
2864
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2865
                    return -1;
2866
                }
2867

    
2868
                if(reordering_of_pic_nums_idc<3){
2869
                    if(reordering_of_pic_nums_idc<2){
2870
                        const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2871
                        int frame_num;
2872

    
2873
                        if(abs_diff_pic_num > h->max_pic_num){
2874
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2875
                            return -1;
2876
                        }
2877

    
2878
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2879
                        else                                pred+= abs_diff_pic_num;
2880
                        pred &= h->max_pic_num - 1;
2881

    
2882
                        frame_num = pic_num_extract(h, pred, &pic_structure);
2883

    
2884
                        for(i= h->short_ref_count-1; i>=0; i--){
2885
                            ref = h->short_ref[i];
2886
                            assert(ref->reference);
2887
                            assert(!ref->long_ref);
2888
                            if(
2889
                                   ref->frame_num == frame_num &&
2890
                                   (ref->reference & pic_structure)
2891
                              )
2892
                                break;
2893
                        }
2894
                        if(i>=0)
2895
                            ref->pic_id= pred;
2896
                    }else{
2897
                        int long_idx;
2898
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2899

    
2900
                        long_idx= pic_num_extract(h, pic_id, &pic_structure);
2901

    
2902
                        if(long_idx>31){
2903
                            av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2904
                            return -1;
2905
                        }
2906
                        ref = h->long_ref[long_idx];
2907
                        assert(!(ref && !ref->reference));
2908
                        if(ref && (ref->reference & pic_structure)){
2909
                            ref->pic_id= pic_id;
2910
                            assert(ref->long_ref);
2911
                            i=0;
2912
                        }else{
2913
                            i=-1;
2914
                        }
2915
                    }
2916

    
2917
                    if (i < 0) {
2918
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2919
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2920
                    } else {
2921
                        for(i=index; i+1<h->ref_count[list]; i++){
2922
                            if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
2923
                                break;
2924
                        }
2925
                        for(; i > index; i--){
2926
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2927
                        }
2928
                        h->ref_list[list][index]= *ref;
2929
                        if (FIELD_PICTURE){
2930
                            pic_as_field(&h->ref_list[list][index], pic_structure);
2931
                        }
2932
                    }
2933
                }else{
2934
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2935
                    return -1;
2936
                }
2937
            }
2938
        }
2939
    }
2940
    for(list=0; list<h->list_count; list++){
2941
        for(index= 0; index < h->ref_count[list]; index++){
2942
            if(!h->ref_list[list][index].data[0]){
2943
                av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
2944
                h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
2945
            }
2946
        }
2947
    }
2948

    
2949
    return 0;
2950
}
2951

    
2952
static void fill_mbaff_ref_list(H264Context *h){
2953
    int list, i, j;
2954
    for(list=0; list<2; list++){ //FIXME try list_count
2955
        for(i=0; i<h->ref_count[list]; i++){
2956
            Picture *frame = &h->ref_list[list][i];
2957
            Picture *field = &h->ref_list[list][16+2*i];
2958
            field[0] = *frame;
2959
            for(j=0; j<3; j++)
2960
                field[0].linesize[j] <<= 1;
2961
            field[0].reference = PICT_TOP_FIELD;
2962
            field[0].poc= field[0].field_poc[0];
2963
            field[1] = field[0];
2964
            for(j=0; j<3; j++)
2965
                field[1].data[j] += frame->linesize[j];
2966
            field[1].reference = PICT_BOTTOM_FIELD;
2967
            field[1].poc= field[1].field_poc[1];
2968

    
2969
            h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
2970
            h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
2971
            for(j=0; j<2; j++){
2972
                h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
2973
                h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
2974
            }
2975
        }
2976
    }
2977
    for(j=0; j<h->ref_count[1]; j++){
2978
        for(i=0; i<h->ref_count[0]; i++)
2979
            h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
2980
        memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));
2981
        memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
2982
    }
2983
}
2984

    
2985
static int pred_weight_table(H264Context *h){
2986
    MpegEncContext * const s = &h->s;
2987
    int list, i;
2988
    int luma_def, chroma_def;
2989

    
2990
    h->use_weight= 0;
2991
    h->use_weight_chroma= 0;
2992
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2993
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2994
    luma_def = 1<<h->luma_log2_weight_denom;
2995
    chroma_def = 1<<h->chroma_log2_weight_denom;
2996

    
2997
    for(list=0; list<2; list++){
2998
        for(i=0; i<h->ref_count[list]; i++){
2999
            int luma_weight_flag, chroma_weight_flag;
3000

    
3001
            luma_weight_flag= get_bits1(&s->gb);
3002
            if(luma_weight_flag){
3003
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3004
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3005
                if(   h->luma_weight[list][i] != luma_def
3006
                   || h->luma_offset[list][i] != 0)
3007
                    h->use_weight= 1;
3008
            }else{
3009
                h->luma_weight[list][i]= luma_def;
3010
                h->luma_offset[list][i]= 0;
3011
            }
3012

    
3013
            if(CHROMA){
3014
                chroma_weight_flag= get_bits1(&s->gb);
3015
                if(chroma_weight_flag){
3016
                    int j;
3017
                    for(j=0; j<2; j++){
3018
                        h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3019
                        h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3020
                        if(   h->chroma_weight[list][i][j] != chroma_def
3021
                        || h->chroma_offset[list][i][j] != 0)
3022
                            h->use_weight_chroma= 1;
3023
                    }
3024
                }else{
3025
                    int j;
3026
                    for(j=0; j<2; j++){
3027
                        h->chroma_weight[list][i][j]= chroma_def;
3028
                        h->chroma_offset[list][i][j]= 0;
3029
                    }
3030
                }
3031
            }
3032
        }
3033
        if(h->slice_type_nos != FF_B_TYPE) break;
3034
    }
3035
    h->use_weight= h->use_weight || h->use_weight_chroma;
3036
    return 0;
3037
}
3038

    
3039
static void implicit_weight_table(H264Context *h){
3040
    MpegEncContext * const s = &h->s;
3041
    int ref0, ref1;
3042
    int cur_poc = s->current_picture_ptr->poc;
3043

    
3044
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3045
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3046
        h->use_weight= 0;
3047
        h->use_weight_chroma= 0;
3048
        return;
3049
    }
3050

    
3051
    h->use_weight= 2;
3052
    h->use_weight_chroma= 2;
3053
    h->luma_log2_weight_denom= 5;
3054
    h->chroma_log2_weight_denom= 5;
3055

    
3056
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3057
        int poc0 = h->ref_list[0][ref0].poc;
3058
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3059
            int poc1 = h->ref_list[1][ref1].poc;
3060
            int td = av_clip(poc1 - poc0, -128, 127);
3061
            if(td){
3062
                int tb = av_clip(cur_poc - poc0, -128, 127);
3063
                int tx = (16384 + (FFABS(td) >> 1)) / td;
3064
                int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3065
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3066
                    h->implicit_weight[ref0][ref1] = 32;
3067
                else
3068
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3069
            }else
3070
                h->implicit_weight[ref0][ref1] = 32;
3071
        }
3072
    }
3073
}
3074

    
3075
/**
3076
 * Mark a picture as no longer needed for reference. The refmask
3077
 * argument allows unreferencing of individual fields or the whole frame.
3078
 * If the picture becomes entirely unreferenced, but is being held for
3079
 * display purposes, it is marked as such.
3080
 * @param refmask mask of fields to unreference; the mask is bitwise
3081
 *                anded with the reference marking of pic
3082
 * @return non-zero if pic becomes entirely unreferenced (except possibly
3083
 *         for display purposes) zero if one of the fields remains in
3084
 *         reference
3085
 */
3086
static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
3087
    int i;
3088
    if (pic->reference &= refmask) {
3089
        return 0;
3090
    } else {
3091
        for(i = 0; h->delayed_pic[i]; i++)
3092
            if(pic == h->delayed_pic[i]){
3093
                pic->reference=DELAYED_PIC_REF;
3094
                break;
3095
            }
3096
        return 1;
3097
    }
3098
}
3099

    
3100
/**
3101
 * instantaneous decoder refresh.
3102
 */
3103
static void idr(H264Context *h){
3104
    int i;
3105

    
3106
    for(i=0; i<16; i++){
3107
        remove_long(h, i, 0);
3108
    }
3109
    assert(h->long_ref_count==0);
3110

    
3111
    for(i=0; i<h->short_ref_count; i++){
3112
        unreference_pic(h, h->short_ref[i], 0);
3113
        h->short_ref[i]= NULL;
3114
    }
3115
    h->short_ref_count=0;
3116
    h->prev_frame_num= 0;
3117
    h->prev_frame_num_offset= 0;
3118
    h->prev_poc_msb=
3119
    h->prev_poc_lsb= 0;
3120
}
3121

    
3122
/* forget old pics after a seek */
3123
static void flush_dpb(AVCodecContext *avctx){
3124
    H264Context *h= avctx->priv_data;
3125
    int i;
3126
    for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
3127
        if(h->delayed_pic[i])
3128
            h->delayed_pic[i]->reference= 0;
3129
        h->delayed_pic[i]= NULL;
3130
    }
3131
    h->outputed_poc= INT_MIN;
3132
    idr(h);
3133
    if(h->s.current_picture_ptr)
3134
        h->s.current_picture_ptr->reference= 0;
3135
    h->s.first_field= 0;
3136
    ff_mpeg_flush(avctx);
3137
}
3138

    
3139
/**
3140
 * Find a Picture in the short term reference list by frame number.
3141
 * @param frame_num frame number to search for
3142
 * @param idx the index into h->short_ref where returned picture is found
3143
 *            undefined if no picture found.
3144
 * @return pointer to the found picture, or NULL if no pic with the provided
3145
 *                 frame number is found
3146
 */
3147
static Picture * find_short(H264Context *h, int frame_num, int *idx){
3148
    MpegEncContext * const s = &h->s;
3149
    int i;
3150

    
3151
    for(i=0; i<h->short_ref_count; i++){
3152
        Picture *pic= h->short_ref[i];
3153
        if(s->avctx->debug&FF_DEBUG_MMCO)
3154
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3155
        if(pic->frame_num == frame_num) {
3156
            *idx = i;
3157
            return pic;
3158
        }
3159
    }
3160
    return NULL;
3161
}
3162

    
3163
/**
3164
 * Remove a picture from the short term reference list by its index in
3165
 * that list.  This does no checking on the provided index; it is assumed
3166
 * to be valid. Other list entries are shifted down.
3167
 * @param i index into h->short_ref of picture to remove.
3168
 */
3169
static void remove_short_at_index(H264Context *h, int i){
3170
    assert(i >= 0 && i < h->short_ref_count);
3171
    h->short_ref[i]= NULL;
3172
    if (--h->short_ref_count)
3173
        memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i)*sizeof(Picture*));
3174
}
3175

    
3176
/**
3177
 *
3178
 * @return the removed picture or NULL if an error occurs
3179
 */
3180
static Picture * remove_short(H264Context *h, int frame_num, int ref_mask){
3181
    MpegEncContext * const s = &h->s;
3182
    Picture *pic;
3183
    int i;
3184

    
3185
    if(s->avctx->debug&FF_DEBUG_MMCO)
3186
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3187

    
3188
    pic = find_short(h, frame_num, &i);
3189
    if (pic){
3190
        if(unreference_pic(h, pic, ref_mask))
3191
        remove_short_at_index(h, i);
3192
    }
3193

    
3194
    return pic;
3195
}
3196

    
3197
/**
3198
 * Remove a picture from the long term reference list by its index in
3199
 * that list.
3200
 * @return the removed picture or NULL if an error occurs
3201
 */
3202
static Picture * remove_long(H264Context *h, int i, int ref_mask){
3203
    Picture *pic;
3204

    
3205
    pic= h->long_ref[i];
3206
    if (pic){
3207
        if(unreference_pic(h, pic, ref_mask)){
3208
            assert(h->long_ref[i]->long_ref == 1);
3209
            h->long_ref[i]->long_ref= 0;
3210
            h->long_ref[i]= NULL;
3211
            h->long_ref_count--;
3212
        }
3213
    }
3214

    
3215
    return pic;
3216
}
3217

    
3218
/**
3219
 * print short term list
3220
 */
3221
static void print_short_term(H264Context *h) {
3222
    uint32_t i;
3223
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3224
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3225
        for(i=0; i<h->short_ref_count; i++){
3226
            Picture *pic= h->short_ref[i];
3227
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3228
        }
3229
    }
3230
}
3231

    
3232
/**
3233
 * print long term list
3234
 */
3235
static void print_long_term(H264Context *h) {
3236
    uint32_t i;
3237
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3238
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3239
        for(i = 0; i < 16; i++){
3240
            Picture *pic= h->long_ref[i];
3241
            if (pic) {
3242
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3243
            }
3244
        }
3245
    }
3246
}
3247

    
3248
/**
3249
 * Executes the reference picture marking (memory management control operations).
3250
 */
3251
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3252
    MpegEncContext * const s = &h->s;
3253
    int i, j;
3254
    int current_ref_assigned=0;
3255
    Picture *pic;
3256

    
3257
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3258
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3259

    
3260
    for(i=0; i<mmco_count; i++){
3261
        int structure, frame_num;
3262
        if(s->avctx->debug&FF_DEBUG_MMCO)
3263
            av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_pic_num, h->mmco[i].long_arg);
3264

    
3265
        if(   mmco[i].opcode == MMCO_SHORT2UNUSED
3266
           || mmco[i].opcode == MMCO_SHORT2LONG){
3267
            frame_num = pic_num_extract(h, mmco[i].short_pic_num, &structure);
3268
            pic = find_short(h, frame_num, &j);
3269
            if(!pic){
3270
                if(mmco[i].opcode != MMCO_SHORT2LONG || !h->long_ref[mmco[i].long_arg]
3271
                   || h->long_ref[mmco[i].long_arg]->frame_num != frame_num)
3272
                av_log(h->s.avctx, AV_LOG_ERROR, "mmco: unref short failure\n");
3273
                continue;
3274
            }
3275
        }
3276

    
3277
        switch(mmco[i].opcode){
3278
        case MMCO_SHORT2UNUSED:
3279
            if(s->avctx->debug&FF_DEBUG_MMCO)
3280
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref short %d count %d\n", h->mmco[i].short_pic_num, h->short_ref_count);
3281
            remove_short(h, frame_num, structure ^ PICT_FRAME);
3282
            break;
3283
        case MMCO_SHORT2LONG:
3284
                if (h->long_ref[mmco[i].long_arg] != pic)
3285
                    remove_long(h, mmco[i].long_arg, 0);
3286

    
3287
                remove_short_at_index(h, j);
3288
                h->long_ref[ mmco[i].long_arg ]= pic;
3289
                if (h->long_ref[ mmco[i].long_arg ]){
3290
                    h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3291
                    h->long_ref_count++;
3292
                }
3293
            break;
3294
        case MMCO_LONG2UNUSED:
3295
            j = pic_num_extract(h, mmco[i].long_arg, &structure);
3296
            pic = h->long_ref[j];
3297
            if (pic) {
3298
                remove_long(h, j, structure ^ PICT_FRAME);
3299
            } else if(s->avctx->debug&FF_DEBUG_MMCO)
3300
                av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: unref long failure\n");
3301
            break;
3302
        case MMCO_LONG:
3303
                    // Comment below left from previous code as it is an interresting note.
3304
                    /* First field in pair is in short term list or
3305
                     * at a different long term index.
3306
                     * This is not allowed; see 7.4.3.3, notes 2 and 3.
3307
                     * Report the problem and keep the pair where it is,
3308
                     * and mark this field valid.
3309
                     */
3310

    
3311
            if (h->long_ref[mmco[i].long_arg] != s->current_picture_ptr) {
3312
                remove_long(h, mmco[i].long_arg, 0);
3313

    
3314
                h->long_ref[ mmco[i].long_arg ]= s->current_picture_ptr;
3315
                h->long_ref[ mmco[i].long_arg ]->long_ref=1;
3316
                h->long_ref_count++;
3317
            }
3318

    
3319
            s->current_picture_ptr->reference |= s->picture_structure;
3320
            current_ref_assigned=1;
3321
            break;
3322
        case MMCO_SET_MAX_LONG:
3323
            assert(mmco[i].long_arg <= 16);
3324
            // just remove the long term which index is greater than new max
3325
            for(j = mmco[i].long_arg; j<16; j++){
3326
                remove_long(h, j, 0);
3327
            }
3328
            break;
3329
        case MMCO_RESET:
3330
            while(h->short_ref_count){
3331
                remove_short(h, h->short_ref[0]->frame_num, 0);
3332
            }
3333
            for(j = 0; j < 16; j++) {
3334
                remove_long(h, j, 0);
3335
            }
3336
            s->current_picture_ptr->poc=
3337
            s->current_picture_ptr->field_poc[0]=
3338
            s->current_picture_ptr->field_poc[1]=
3339
            h->poc_lsb=
3340
            h->poc_msb=
3341
            h->frame_num=
3342
            s->current_picture_ptr->frame_num= 0;
3343
            break;
3344
        default: assert(0);
3345
        }
3346
    }
3347

    
3348
    if (!current_ref_assigned) {
3349
        /* Second field of complementary field pair; the first field of
3350
         * which is already referenced. If short referenced, it
3351
         * should be first entry in short_ref. If not, it must exist
3352
         * in long_ref; trying to put it on the short list here is an
3353
         * error in the encoded bit stream (ref: 7.4.3.3, NOTE 2 and 3).
3354
         */
3355
        if (h->short_ref_count && h->short_ref[0] == s->current_picture_ptr) {
3356
            /* Just mark the second field valid */
3357
            s->current_picture_ptr->reference = PICT_FRAME;
3358
        } else if (s->current_picture_ptr->long_ref) {
3359
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term reference "
3360
                                             "assignment for second field "
3361
                                             "in complementary field pair "
3362
                                             "(first field is long term)\n");
3363
        } else {
3364
            pic= remove_short(h, s->current_picture_ptr->frame_num, 0);
3365
            if(pic){
3366
                av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3367
            }
3368

    
3369
            if(h->short_ref_count)
3370
                memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3371

    
3372
            h->short_ref[0]= s->current_picture_ptr;
3373
            h->short_ref_count++;
3374
            s->current_picture_ptr->reference |= s->picture_structure;
3375
        }
3376
    }
3377

    
3378
    if (h->long_ref_count + h->short_ref_count > h->sps.ref_frame_count){
3379

    
3380
        /* We have too many reference frames, probably due to corrupted
3381
         * stream. Need to discard one frame. Prevents overrun of the
3382
         * short_ref and long_ref buffers.
3383
         */
3384
        av_log(h->s.avctx, AV_LOG_ERROR,
3385
               "number of reference frames exceeds max (probably "
3386
               "corrupt input), discarding one\n");
3387

    
3388
        if (h->long_ref_count && !h->short_ref_count) {
3389
            for (i = 0; i < 16; ++i)
3390
                if (h->long_ref[i])
3391
                    break;
3392

    
3393
            assert(i < 16);
3394
            remove_long(h, i, 0);
3395
        } else {
3396
            pic = h->short_ref[h->short_ref_count - 1];
3397
            remove_short(h, pic->frame_num, 0);
3398
        }
3399
    }
3400

    
3401
    print_short_term(h);
3402
    print_long_term(h);
3403
    return 0;
3404
}
3405

    
3406
static int decode_ref_pic_marking(H264Context *h, GetBitContext *gb){
3407
    MpegEncContext * const s = &h->s;
3408
    int i;
3409

    
3410
    h->mmco_index= 0;
3411
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3412
        s->broken_link= get_bits1(gb) -1;
3413
        if(get_bits1(gb)){
3414
            h->mmco[0].opcode= MMCO_LONG;
3415
            h->mmco[0].long_arg= 0;
3416
            h->mmco_index= 1;
3417
        }
3418
    }else{
3419
        if(get_bits1(gb)){ // adaptive_ref_pic_marking_mode_flag
3420
            for(i= 0; i<MAX_MMCO_COUNT; i++) {
3421
                MMCOOpcode opcode= get_ue_golomb_31(gb);
3422

    
3423
                h->mmco[i].opcode= opcode;
3424
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3425
                    h->mmco[i].short_pic_num= (h->curr_pic_num - get_ue_golomb(gb) - 1) & (h->max_pic_num - 1);
3426
/*                    if(h->mmco[i].short_pic_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_pic_num ] == NULL){
3427
                        av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);
3428
                        return -1;
3429
                    }*/
3430
                }
3431
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3432
                    unsigned int long_arg= get_ue_golomb_31(gb);
3433
                    if(long_arg >= 32 || (long_arg >= 16 && !(opcode == MMCO_LONG2UNUSED && FIELD_PICTURE))){
3434
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3435
                        return -1;
3436
                    }
3437
                    h->mmco[i].long_arg= long_arg;
3438
                }
3439

    
3440
                if(opcode > (unsigned)MMCO_LONG){
3441
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3442
                    return -1;
3443
                }
3444
                if(opcode == MMCO_END)
3445
                    break;
3446
            }
3447
            h->mmco_index= i;
3448
        }else{
3449
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3450

    
3451
            if(h->short_ref_count && h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count &&
3452
                    !(FIELD_PICTURE && !s->first_field && s->current_picture_ptr->reference)) {
3453
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3454
                h->mmco[0].short_pic_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3455
                h->mmco_index= 1;
3456
                if (FIELD_PICTURE) {
3457
                    h->mmco[0].short_pic_num *= 2;
3458
                    h->mmco[1].opcode= MMCO_SHORT2UNUSED;
3459
                    h->mmco[1].short_pic_num= h->mmco[0].short_pic_num + 1;
3460
                    h->mmco_index= 2;
3461
                }
3462
            }
3463
        }
3464
    }
3465

    
3466
    return 0;
3467
}
3468

    
3469
static int init_poc(H264Context *h){
3470
    MpegEncContext * const s = &h->s;
3471
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3472
    int field_poc[2];
3473
    Picture *cur = s->current_picture_ptr;
3474

    
3475
    h->frame_num_offset= h->prev_frame_num_offset;
3476
    if(h->frame_num < h->prev_frame_num)
3477
        h->frame_num_offset += max_frame_num;
3478

    
3479
    if(h->sps.poc_type==0){
3480
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3481

    
3482
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3483
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3484
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3485
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3486
        else
3487
            h->poc_msb = h->prev_poc_msb;
3488
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3489
        field_poc[0] =
3490
        field_poc[1] = h->poc_msb + h->poc_lsb;
3491
        if(s->picture_structure == PICT_FRAME)
3492
            field_poc[1] += h->delta_poc_bottom;
3493
    }else if(h->sps.poc_type==1){
3494
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3495
        int i;
3496

    
3497
        if(h->sps.poc_cycle_length != 0)
3498
            abs_frame_num = h->frame_num_offset + h->frame_num;
3499
        else
3500
            abs_frame_num = 0;
3501

    
3502
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3503
            abs_frame_num--;
3504

    
3505
        expected_delta_per_poc_cycle = 0;
3506
        for(i=0; i < h->sps.poc_cycle_length; i++)
3507
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3508

    
3509
        if(abs_frame_num > 0){
3510
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3511
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3512

    
3513
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3514
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3515
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3516
        } else
3517
            expectedpoc = 0;
3518

    
3519
        if(h->nal_ref_idc == 0)
3520
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3521

    
3522
        field_poc[0] = expectedpoc + h->delta_poc[0];
3523
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3524

    
3525
        if(s->picture_structure == PICT_FRAME)
3526
            field_poc[1] += h->delta_poc[1];
3527
    }else{
3528
        int poc= 2*(h->frame_num_offset + h->frame_num);
3529

    
3530
        if(!h->nal_ref_idc)
3531
            poc--;
3532

    
3533
        field_poc[0]= poc;
3534
        field_poc[1]= poc;
3535
    }
3536

    
3537
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3538
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3539
    if(s->picture_structure != PICT_TOP_FIELD)
3540
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3541
    cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
3542

    
3543
    return 0;
3544
}
3545

    
3546

    
3547
/**
3548
 * initialize scan tables
3549
 */
3550
static void init_scan_tables(H264Context *h){
3551
    MpegEncContext * const s = &h->s;
3552
    int i;
3553
    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
3554
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
3555
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
3556
    }else{
3557
        for(i=0; i<16; i++){
3558
#define T(x) (x>>2) | ((x<<2) & 0xF)
3559
            h->zigzag_scan[i] = T(zigzag_scan[i]);
3560
            h-> field_scan[i] = T( field_scan[i]);
3561
#undef T
3562
        }
3563
    }
3564
    if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){
3565
        memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));
3566
        memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));
3567
        memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));
3568
        memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));
3569
    }else{
3570
        for(i=0; i<64; i++){
3571
#define T(x) (x>>3) | ((x&7)<<3)
3572
            h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);
3573
            h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
3574
            h->field_scan8x8[i]        = T(field_scan8x8[i]);
3575
            h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
3576
#undef T
3577
        }
3578
    }
3579
    if(h->sps.transform_bypass){ //FIXME same ugly
3580
        h->zigzag_scan_q0          = zigzag_scan;
3581
        h->zigzag_scan8x8_q0       = zigzag_scan8x8;
3582
        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
3583
        h->field_scan_q0           = field_scan;
3584
        h->field_scan8x8_q0        = field_scan8x8;
3585
        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
3586
    }else{
3587
        h->zigzag_scan_q0          = h->zigzag_scan;
3588
        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
3589
        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
3590
        h->field_scan_q0           = h->field_scan;
3591
        h->field_scan8x8_q0        = h->field_scan8x8;
3592
        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
3593
    }
3594
}
3595

    
3596
/**
3597
 * Replicates H264 "master" context to thread contexts.
3598
 */
3599
static void clone_slice(H264Context *dst, H264Context *src)
3600
{
3601
    memcpy(dst->block_offset,     src->block_offset, sizeof(dst->block_offset));
3602
    dst->s.current_picture_ptr  = src->s.current_picture_ptr;
3603
    dst->s.current_picture      = src->s.current_picture;
3604
    dst->s.linesize             = src->s.linesize;
3605
    dst->s.uvlinesize           = src->s.uvlinesize;
3606
    dst->s.first_field          = src->s.first_field;
3607

    
3608
    dst->prev_poc_msb           = src->prev_poc_msb;
3609
    dst->prev_poc_lsb           = src->prev_poc_lsb;
3610
    dst->prev_frame_num_offset  = src->prev_frame_num_offset;
3611
    dst->prev_frame_num         = src->prev_frame_num;
3612
    dst->short_ref_count        = src->short_ref_count;
3613

    
3614
    memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
3615
    memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
3616
    memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3617
    memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
3618

    
3619
    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
3620
    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
3621
}
3622

    
3623
/**
3624
 * decodes a slice header.
3625
 * This will also call MPV_common_init() and frame_start() as needed.
3626
 *
3627
 * @param h h264context
3628
 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
3629
 *
3630
 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3631
 */
3632
static int decode_slice_header(H264Context *h, H264Context *h0){
3633
    MpegEncContext * const s = &h->s;
3634
    MpegEncContext * const s0 = &h0->s;
3635
    unsigned int first_mb_in_slice;
3636
    unsigned int pps_id;
3637
    int num_ref_idx_active_override_flag;
3638
    unsigned int slice_type, tmp, i, j;
3639
    int default_ref_list_done = 0;
3640
    int last_pic_structure;
3641

    
3642
    s->dropable= h->nal_ref_idc == 0;
3643

    
3644
    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
3645
        s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
3646
        s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
3647
    }else{
3648
        s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
3649
        s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
3650
    }
3651

    
3652
    first_mb_in_slice= get_ue_golomb(&s->gb);
3653

    
3654
    if((s->flags2 & CODEC_FLAG2_CHUNKS) && first_mb_in_slice == 0){
3655
        h0->current_slice = 0;
3656
        if (!s0->first_field)
3657
            s->current_picture_ptr= NULL;
3658
    }
3659

    
3660
    slice_type= get_ue_golomb_31(&s->gb);
3661
    if(slice_type > 9){
3662
        av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
3663
        return -1;
3664
    }
3665
    if(slice_type > 4){
3666
        slice_type -= 5;
3667
        h->slice_type_fixed=1;
3668
    }else
3669
        h->slice_type_fixed=0;
3670

    
3671
    slice_type= golomb_to_pict_type[ slice_type ];
3672
    if (slice_type == FF_I_TYPE
3673
        || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
3674
        default_ref_list_done = 1;
3675
    }
3676
    h->slice_type= slice_type;
3677
    h->slice_type_nos= slice_type & 3;
3678

    
3679
    s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
3680
    if (s->pict_type == FF_B_TYPE && s0->last_picture_ptr == NULL) {
3681
        av_log(h->s.avctx, AV_LOG_ERROR,
3682
               "B picture before any references, skipping\n");
3683
        return -1;
3684
    }
3685

    
3686
    pps_id= get_ue_golomb(&s->gb);
3687
    if(pps_id>=MAX_PPS_COUNT){
3688
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3689
        return -1;
3690
    }
3691
    if(!h0->pps_buffers[pps_id]) {
3692
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
3693
        return -1;
3694
    }
3695
    h->pps= *h0->pps_buffers[pps_id];
3696

    
3697
    if(!h0->sps_buffers[h->pps.sps_id]) {
3698
        av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
3699
        return -1;
3700
    }
3701
    h->sps = *h0->sps_buffers[h->pps.sps_id];
3702

    
3703
    if(h == h0 && h->dequant_coeff_pps != pps_id){
3704
        h->dequant_coeff_pps = pps_id;
3705
        init_dequant_tables(h);
3706
    }
3707

    
3708
    s->mb_width= h->sps.mb_width;
3709
    s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3710

    
3711
    h->b_stride=  s->mb_width*4;
3712
    h->b8_stride= s->mb_width*2;
3713

    
3714
    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
3715
    if(h->sps.frame_mbs_only_flag)
3716
        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
3717
    else
3718
        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 3);
3719

    
3720
    if (s->context_initialized
3721
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3722
        if(h != h0)
3723
            return -1;   // width / height changed during parallelized decoding
3724
        free_tables(h);
3725
        flush_dpb(s->avctx);
3726
        MPV_common_end(s);
3727
    }
3728
    if (!s->context_initialized) {
3729
        if(h != h0)
3730
            return -1;  // we cant (re-)initialize context during parallel decoding
3731
        if (MPV_common_init(s) < 0)
3732
            return -1;
3733
        s->first_field = 0;
3734

    
3735
        init_scan_tables(h);
3736
        alloc_tables(h);
3737

    
3738
        for(i = 1; i < s->avctx->thread_count; i++) {
3739
            H264Context *c;
3740
            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
3741
            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
3742
            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
3743
            c->sps = h->sps;
3744
            c->pps = h->pps;
3745
            init_scan_tables(c);
3746
            clone_tables(c, h);
3747
        }
3748

    
3749
        for(i = 0; i < s->avctx->thread_count; i++)
3750
            if(context_init(h->thread_context[i]) < 0)
3751
                return -1;
3752

    
3753
        s->avctx->width = s->width;
3754
        s->avctx->height = s->height;
3755
        s->avctx->sample_aspect_ratio= h->sps.sar;
3756
        if(!s->avctx->sample_aspect_ratio.den)
3757
            s->avctx->sample_aspect_ratio.den = 1;
3758

    
3759
        if(h->sps.timing_info_present_flag){
3760
            s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};
3761
            if(h->x264_build > 0 && h->x264_build < 44)
3762
                s->avctx->time_base.den *= 2;
3763
            av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3764
                      s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);
3765
        }
3766
    }
3767

    
3768
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3769

    
3770
    h->mb_mbaff = 0;
3771
    h->mb_aff_frame = 0;
3772
    last_pic_structure = s0->picture_structure;
3773
    if(h->sps.frame_mbs_only_flag){
3774
        s->picture_structure= PICT_FRAME;
3775
    }else{
3776
        if(get_bits1(&s->gb)) { //field_pic_flag
3777
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3778
        } else {
3779
            s->picture_structure= PICT_FRAME;
3780
            h->mb_aff_frame = h->sps.mb_aff;
3781
        }
3782
    }
3783
    h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
3784

    
3785
    if(h0->current_slice == 0){
3786
        while(h->frame_num !=  h->prev_frame_num &&
3787
              h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
3788
            av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
3789
            frame_start(h);
3790
            h->prev_frame_num++;
3791
            h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
3792
            s->current_picture_ptr->frame_num= h->prev_frame_num;
3793
            execute_ref_pic_marking(h, NULL, 0);
3794
        }
3795

    
3796
        /* See if we have a decoded first field looking for a pair... */
3797
        if (s0->first_field) {
3798
            assert(s0->current_picture_ptr);
3799
            assert(s0->current_picture_ptr->data[0]);
3800
            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
3801

    
3802
            /* figure out if we have a complementary field pair */
3803
            if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3804
                /*
3805
                 * Previous field is unmatched. Don't display it, but let it
3806
                 * remain for reference if marked as such.
3807
                 */
3808
                s0->current_picture_ptr = NULL;
3809
                s0->first_field = FIELD_PICTURE;
3810

    
3811
            } else {
3812
                if (h->nal_ref_idc &&
3813
                        s0->current_picture_ptr->reference &&
3814
                        s0->current_picture_ptr->frame_num != h->frame_num) {
3815
                    /*
3816
                     * This and previous field were reference, but had
3817
                     * different frame_nums. Consider this field first in
3818
                     * pair. Throw away previous field except for reference
3819
                     * purposes.
3820
                     */
3821
                    s0->first_field = 1;
3822
                    s0->current_picture_ptr = NULL;
3823

    
3824
                } else {
3825
                    /* Second field in complementary pair */
3826
                    s0->first_field = 0;
3827
                }
3828
            }
3829

    
3830
        } else {
3831
            /* Frame or first field in a potentially complementary pair */
3832
            assert(!s0->current_picture_ptr);
3833
            s0->first_field = FIELD_PICTURE;
3834
        }
3835

    
3836
        if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) {
3837
            s0->first_field = 0;
3838
            return -1;
3839
        }
3840
    }
3841
    if(h != h0)
3842
        clone_slice(h, h0);
3843

    
3844
    s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3845

    
3846
    assert(s->mb_num == s->mb_width * s->mb_height);
3847
    if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3848
       first_mb_in_slice                    >= s->mb_num){
3849
        av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3850
        return -1;
3851
    }
3852
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3853
    s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3854
    if (s->picture_structure == PICT_BOTTOM_FIELD)
3855
        s->resync_mb_y = s->mb_y = s->mb_y + 1;
3856
    assert(s->mb_y < s->mb_height);
3857

    
3858
    if(s->picture_structure==PICT_FRAME){
3859
        h->curr_pic_num=   h->frame_num;
3860
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3861
    }else{
3862
        h->curr_pic_num= 2*h->frame_num + 1;
3863
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3864
    }
3865

    
3866
    if(h->nal_unit_type == NAL_IDR_SLICE){
3867
        get_ue_golomb(&s->gb); /* idr_pic_id */
3868
    }
3869

    
3870
    if(h->sps.poc_type==0){
3871
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3872

    
3873
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3874
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3875
        }
3876
    }
3877

    
3878
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3879
        h->delta_poc[0]= get_se_golomb(&s->gb);
3880

    
3881
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3882
            h->delta_poc[1]= get_se_golomb(&s->gb);
3883
    }
3884

    
3885
    init_poc(h);
3886

    
3887
    if(h->pps.redundant_pic_cnt_present){
3888
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3889
    }
3890

    
3891
    //set defaults, might be overridden a few lines later
3892
    h->ref_count[0]= h->pps.ref_count[0];
3893
    h->ref_count[1]= h->pps.ref_count[1];
3894

    
3895
    if(h->slice_type_nos != FF_I_TYPE){
3896
        if(h->slice_type_nos == FF_B_TYPE){
3897
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3898
        }
3899
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3900

    
3901
        if(num_ref_idx_active_override_flag){
3902
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3903
            if(h->slice_type_nos==FF_B_TYPE)
3904
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3905

    
3906
            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
3907
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3908
                h->ref_count[0]= h->ref_count[1]= 1;
3909
                return -1;
3910
            }
3911
        }
3912
        if(h->slice_type_nos == FF_B_TYPE)
3913
            h->list_count= 2;
3914
        else
3915
            h->list_count= 1;
3916
    }else
3917
        h->list_count= 0;
3918

    
3919
    if(!default_ref_list_done){
3920
        fill_default_ref_list(h);
3921
    }
3922

    
3923
    if(h->slice_type_nos!=FF_I_TYPE && decode_ref_pic_list_reordering(h) < 0)
3924
        return -1;
3925

    
3926
    if(h->slice_type_nos!=FF_I_TYPE){
3927
        s->last_picture_ptr= &h->ref_list[0][0];
3928
        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3929
    }
3930
    if(h->slice_type_nos==FF_B_TYPE){
3931
        s->next_picture_ptr= &h->ref_list[1][0];
3932
        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3933
    }
3934

    
3935
    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
3936
       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
3937
        pred_weight_table(h);
3938
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE)
3939
        implicit_weight_table(h);
3940
    else
3941
        h->use_weight = 0;
3942

    
3943
    if(h->nal_ref_idc)
3944
        decode_ref_pic_marking(h0, &s->gb);
3945

    
3946
    if(FRAME_MBAFF)
3947
        fill_mbaff_ref_list(h);
3948

    
3949
    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
3950
        direct_dist_scale_factor(h);
3951
    direct_ref_list_init(h);
3952

    
3953
    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
3954
        tmp = get_ue_golomb_31(&s->gb);
3955
        if(tmp > 2){
3956
            av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3957
            return -1;
3958
        }
3959
        h->cabac_init_idc= tmp;
3960
    }
3961

    
3962
    h->last_qscale_diff = 0;
3963
    tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3964
    if(tmp>51){
3965
        av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3966
        return -1;
3967
    }
3968
    s->qscale= tmp;
3969
    h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3970
    h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3971
    //FIXME qscale / qp ... stuff
3972
    if(h->slice_type == FF_SP_TYPE){
3973
        get_bits1(&s->gb); /* sp_for_switch_flag */
3974
    }
3975
    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
3976
        get_se_golomb(&s->gb); /* slice_qs_delta */
3977
    }
3978

    
3979
    h->deblocking_filter = 1;
3980
    h->slice_alpha_c0_offset = 0;
3981
    h->slice_beta_offset = 0;
3982
    if( h->pps.deblocking_filter_parameters_present ) {
3983
        tmp= get_ue_golomb_31(&s->gb);
3984
        if(tmp > 2){
3985
            av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3986
            return -1;
3987
        }
3988
        h->deblocking_filter= tmp;
3989
        if(h->deblocking_filter < 2)
3990
            h->deblocking_filter^= 1; // 1<->0
3991

    
3992
        if( h->deblocking_filter ) {
3993
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3994
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3995
        }
3996
    }
3997

    
3998
    if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
3999
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
4000
       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
4001
       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
4002
        h->deblocking_filter= 0;
4003

    
4004
    if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
4005
        if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
4006
            /* Cheat slightly for speed:
4007
               Do not bother to deblock across slices. */
4008
            h->deblocking_filter = 2;
4009
        } else {
4010
            h0->max_contexts = 1;
4011
            if(!h0->single_decode_warning) {
4012
                av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
4013
                h0->single_decode_warning = 1;
4014
            }
4015
            if(h != h0)
4016
                return 1; // deblocking switched inside frame
4017
        }
4018
    }
4019

    
4020
#if 0 //FMO
4021
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4022
        slice_group_change_cycle= get_bits(&s->gb, ?);
4023
#endif
4024

    
4025
    h0->last_slice_type = slice_type;
4026
    h->slice_num = ++h0->current_slice;
4027
    if(h->slice_num >= MAX_SLICES){
4028
        av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
4029
    }
4030

    
4031
    for(j=0; j<2; j++){
4032
        int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
4033
        ref2frm[0]=
4034
        ref2frm[1]= -1;
4035
        for(i=0; i<16; i++)
4036
            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
4037
                          +(h->ref_list[j][i].reference&3);
4038
        ref2frm[18+0]=
4039
        ref2frm[18+1]= -1;
4040
        for(i=16; i<48; i++)
4041
            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
4042
                          +(h->ref_list[j][i].reference&3);
4043
    }
4044

    
4045
    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
4046
    h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
4047

    
4048
    s->avctx->refs= h->sps.ref_frame_count;
4049

    
4050
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4051
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
4052
               h->slice_num,
4053
               (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4054
               first_mb_in_slice,
4055
               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4056
               pps_id, h->frame_num,
4057
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4058
               h->ref_count[0], h->ref_count[1],
4059
               s->qscale,
4060
               h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4061
               h->use_weight,
4062
               h->use_weight==1 && h->use_weight_chroma ? "c" : "",
4063
               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
4064
               );
4065
    }
4066

    
4067
    return 0;
4068
}
4069

    
4070
/**
4071
 *
4072
 */
4073
static inline int get_level_prefix(GetBitContext *gb){
4074
    unsigned int buf;
4075
    int log;
4076

    
4077
    OPEN_READER(re, gb);
4078
    UPDATE_CACHE(re, gb);
4079
    buf=GET_CACHE(re, gb);
4080

    
4081
    log= 32 - av_log2(buf);
4082
#ifdef TRACE
4083
    print_bin(buf>>(32-log), log);
4084
    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
4085
#endif
4086

    
4087
    LAST_SKIP_BITS(re, gb, log);
4088
    CLOSE_READER(re, gb);
4089

    
4090
    return log-1;
4091
}
4092

    
4093
static inline int get_dct8x8_allowed(H264Context *h){
4094
    if(h->sps.direct_8x8_inference_flag)
4095
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
4096
    else
4097
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
4098
}
4099

    
4100
/**
4101
 * decodes a residual block.
4102
 * @param n block index
4103
 * @param scantable scantable
4104
 * @param max_coeff number of coefficients in the block
4105
 * @return <0 if an error occurred
4106
 */
4107
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){
4108
    MpegEncContext * const s = &h->s;
4109
    static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
4110
    int level[16];
4111
    int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;
4112

    
4113
    //FIXME put trailing_onex into the context
4114

    
4115
    if(n == CHROMA_DC_BLOCK_INDEX){
4116
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4117
        total_coeff= coeff_token>>2;
4118
    }else{
4119
        if(n == LUMA_DC_BLOCK_INDEX){
4120
            total_coeff= pred_non_zero_count(h, 0);
4121
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4122
            total_coeff= coeff_token>>2;
4123
        }else{
4124
            total_coeff= pred_non_zero_count(h, n);
4125
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4126
            total_coeff= coeff_token>>2;
4127
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4128
        }
4129
    }
4130

    
4131
    //FIXME set last_non_zero?
4132

    
4133
    if(total_coeff==0)
4134
        return 0;
4135
    if(total_coeff > (unsigned)max_coeff) {
4136
        av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", s->mb_x, s->mb_y, total_coeff);
4137
        return -1;
4138
    }
4139

    
4140
    trailing_ones= coeff_token&3;
4141
    tprintf(h->s.avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
4142
    assert(total_coeff<=16);
4143

    
4144
    i = show_bits(gb, 3);
4145
    skip_bits(gb, trailing_ones);
4146
    level[0] = 1-((i&4)>>1);
4147
    level[1] = 1-((i&2)   );
4148
    level[2] = 1-((i&1)<<1);
4149

    
4150
    if(trailing_ones<total_coeff) {
4151
        int mask, prefix;
4152
        int suffix_length = total_coeff > 10 && trailing_ones < 3;
4153
        int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4154
        int level_code= cavlc_level_tab[suffix_length][bitsi][0];
4155

    
4156
        skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4157
        if(level_code >= 100){
4158
            prefix= level_code - 100;
4159
            if(prefix == LEVEL_TAB_BITS)
4160
                prefix += get_level_prefix(gb);
4161

    
4162
            //first coefficient has suffix_length equal to 0 or 1
4163
            if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4164
                if(suffix_length)
4165
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4166
                else
4167
                    level_code= (prefix<<suffix_length); //part
4168
            }else if(prefix==14){
4169
                if(suffix_length)
4170
                    level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4171
                else
4172
                    level_code= prefix + get_bits(gb, 4); //part
4173
            }else{
4174
                level_code= (15<<suffix_length) + get_bits(gb, prefix-3); //part
4175
                if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4176
                if(prefix>=16)
4177
                    level_code += (1<<(prefix-3))-4096;
4178
            }
4179

    
4180
            if(trailing_ones < 3) level_code += 2;
4181

    
4182
            suffix_length = 2;
4183
            mask= -(level_code&1);
4184
            level[trailing_ones]= (((2+level_code)>>1) ^ mask) - mask;
4185
        }else{
4186
            if(trailing_ones < 3) level_code += (level_code>>31)|1;
4187

    
4188
            suffix_length = 1;
4189
            if(level_code + 3U > 6U)
4190
                suffix_length++;
4191
            level[trailing_ones]= level_code;
4192
        }
4193

    
4194
        //remaining coefficients have suffix_length > 0
4195
        for(i=trailing_ones+1;i<total_coeff;i++) {
4196
            static const unsigned int suffix_limit[7] = {0,3,6,12,24,48,INT_MAX };
4197
            int bitsi= show_bits(gb, LEVEL_TAB_BITS);
4198
            level_code= cavlc_level_tab[suffix_length][bitsi][0];
4199

    
4200
            skip_bits(gb, cavlc_level_tab[suffix_length][bitsi][1]);
4201
            if(level_code >= 100){
4202
                prefix= level_code - 100;
4203
                if(prefix == LEVEL_TAB_BITS){
4204
                    prefix += get_level_prefix(gb);
4205
                }
4206
                if(prefix<15){
4207
                    level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);
4208
                }else{
4209
                    level_code = (15<<suffix_length) + get_bits(gb, prefix-3);
4210
                    if(prefix>=16)
4211
                        level_code += (1<<(prefix-3))-4096;
4212
                }
4213
                mask= -(level_code&1);
4214
                level_code= (((2+level_code)>>1) ^ mask) - mask;
4215
            }
4216
            level[i]= level_code;
4217

    
4218
            if(suffix_limit[suffix_length] + level_code > 2U*suffix_limit[suffix_length])
4219
                suffix_length++;
4220
        }
4221
    }
4222

    
4223
    if(total_coeff == max_coeff)
4224
        zeros_left=0;
4225
    else{
4226
        if(n == CHROMA_DC_BLOCK_INDEX)
4227
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4228
        else
4229
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4230
    }
4231

    
4232
    coeff_num = zeros_left + total_coeff - 1;
4233
    j = scantable[coeff_num];
4234
    if(n > 24){
4235
        block[j] = level[0];
4236
        for(i=1;i<total_coeff;i++) {
4237
            if(zeros_left <= 0)
4238
                run_before = 0;
4239
            else if(zeros_left < 7){
4240
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4241
            }else{
4242
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4243
            }
4244
            zeros_left -= run_before;
4245
            coeff_num -= 1 + run_before;
4246
            j= scantable[ coeff_num ];
4247

    
4248
            block[j]= level[i];
4249
        }
4250
    }else{
4251
        block[j] = (level[0] * qmul[j] + 32)>>6;
4252
        for(i=1;i<total_coeff;i++) {
4253
            if(zeros_left <= 0)
4254
                run_before = 0;
4255
            else if(zeros_left < 7){
4256
                run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4257
            }else{
4258
                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4259
            }
4260
            zeros_left -= run_before;
4261
            coeff_num -= 1 + run_before;
4262
            j= scantable[ coeff_num ];
4263

    
4264
            block[j]= (level[i] * qmul[j] + 32)>>6;
4265
        }
4266
    }
4267

    
4268
    if(zeros_left<0){
4269
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4270
        return -1;
4271
    }
4272

    
4273
    return 0;
4274
}
4275

    
4276
static void predict_field_decoding_flag(H264Context *h){
4277
    MpegEncContext * const s = &h->s;
4278
    const int mb_xy= h->mb_xy;
4279
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
4280
                ? s->current_picture.mb_type[mb_xy-1]
4281
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
4282
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
4283
                : 0;
4284
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4285
}
4286

    
4287
/**
4288
 * decodes a P_SKIP or B_SKIP macroblock
4289
 */
4290
static void decode_mb_skip(H264Context *h){
4291
    MpegEncContext * const s = &h->s;
4292
    const int mb_xy= h->mb_xy;
4293
    int mb_type=0;
4294

    
4295
    memset(h->non_zero_count[mb_xy], 0, 16);
4296
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4297

    
4298
    if(MB_FIELD)
4299
        mb_type|= MB_TYPE_INTERLACED;
4300

    
4301
    if( h->slice_type_nos == FF_B_TYPE )
4302
    {
4303
        // just for fill_caches. pred_direct_motion will set the real mb_type
4304
        mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4305

    
4306
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4307
        pred_direct_motion(h, &mb_type);
4308
        mb_type|= MB_TYPE_SKIP;
4309
    }
4310
    else
4311
    {
4312
        int mx, my;
4313
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4314

    
4315
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4316
        pred_pskip_motion(h, &mx, &my);
4317
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4318
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4319
    }
4320

    
4321
    write_back_motion(h, mb_type);
4322
    s->current_picture.mb_type[mb_xy]= mb_type;
4323
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4324
    h->slice_table[ mb_xy ]= h->slice_num;
4325
    h->prev_mb_skipped= 1;
4326
}
4327

    
4328
/**
4329
 * decodes a macroblock
4330
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4331
 */
4332
static int decode_mb_cavlc(H264Context *h){
4333
    MpegEncContext * const s = &h->s;
4334
    int mb_xy;
4335
    int partition_count;
4336
    unsigned int mb_type, cbp;
4337
    int dct8x8_allowed= h->pps.transform_8x8_mode;
4338

    
4339
    mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4340

    
4341
    tprintf(s->avctx, "pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4342
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4343
                down the code */
4344
    if(h->slice_type_nos != FF_I_TYPE){
4345
        if(s->mb_skip_run==-1)
4346
            s->mb_skip_run= get_ue_golomb(&s->gb);
4347

    
4348
        if (s->mb_skip_run--) {
4349
            if(FRAME_MBAFF && (s->mb_y&1) == 0){
4350
                if(s->mb_skip_run==0)
4351
                    h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4352
                else
4353
                    predict_field_decoding_flag(h);
4354
            }
4355
            decode_mb_skip(h);
4356
            return 0;
4357
        }
4358
    }
4359
    if(FRAME_MBAFF){
4360
        if( (s->mb_y&1) == 0 )
4361
            h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
4362
    }
4363

    
4364
    h->prev_mb_skipped= 0;
4365

    
4366
    mb_type= get_ue_golomb(&s->gb);
4367
    if(h->slice_type_nos == FF_B_TYPE){
4368
        if(mb_type < 23){
4369
            partition_count= b_mb_type_info[mb_type].partition_count;
4370
            mb_type=         b_mb_type_info[mb_type].type;
4371
        }else{
4372
            mb_type -= 23;
4373
            goto decode_intra_mb;
4374
        }
4375
    }else if(h->slice_type_nos == FF_P_TYPE){
4376
        if(mb_type < 5){
4377
            partition_count= p_mb_type_info[mb_type].partition_count;
4378
            mb_type=         p_mb_type_info[mb_type].type;
4379
        }else{
4380
            mb_type -= 5;
4381
            goto decode_intra_mb;
4382
        }
4383
    }else{
4384
       assert(h->slice_type_nos == FF_I_TYPE);
4385
        if(h->slice_type == FF_SI_TYPE && mb_type)
4386
            mb_type--;
4387
decode_intra_mb:
4388
        if(mb_type > 25){
4389
            av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
4390
            return -1;
4391
        }
4392
        partition_count=0;
4393
        cbp= i_mb_type_info[mb_type].cbp;
4394
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4395
        mb_type= i_mb_type_info[mb_type].type;
4396
    }
4397

    
4398
    if(MB_FIELD)
4399
        mb_type |= MB_TYPE_INTERLACED;
4400

    
4401
    h->slice_table[ mb_xy ]= h->slice_num;
4402

    
4403
    if(IS_INTRA_PCM(mb_type)){
4404
        unsigned int x;
4405

    
4406
        // We assume these blocks are very rare so we do not optimize it.
4407
        align_get_bits(&s->gb);
4408

    
4409
        // The pixels are stored in the same order as levels in h->mb array.
4410
        for(x=0; x < (CHROMA ? 384 : 256); x++){
4411
            ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
4412
        }
4413

    
4414
        // In deblocking, the quantizer is 0
4415
        s->current_picture.qscale_table[mb_xy]= 0;
4416
        // All coeffs are present
4417
        memset(h->non_zero_count[mb_xy], 16, 16);
4418

    
4419
        s->current_picture.mb_type[mb_xy]= mb_type;
4420
        return 0;
4421
    }
4422

    
4423
    if(MB_MBAFF){
4424
        h->ref_count[0] <<= 1;
4425
        h->ref_count[1] <<= 1;
4426
    }
4427

    
4428
    fill_caches(h, mb_type, 0);
4429

    
4430
    //mb_pred
4431
    if(IS_INTRA(mb_type)){
4432
        int pred_mode;
4433
//            init_top_left_availability(h);
4434
        if(IS_INTRA4x4(mb_type)){
4435
            int i;
4436
            int di = 1;
4437
            if(dct8x8_allowed && get_bits1(&s->gb)){
4438
                mb_type |= MB_TYPE_8x8DCT;
4439
                di = 4;
4440
            }
4441

    
4442
//                fill_intra4x4_pred_table(h);
4443
            for(i=0; i<16; i+=di){
4444
                int mode= pred_intra_mode(h, i);
4445

    
4446
                if(!get_bits1(&s->gb)){
4447
                    const int rem_mode= get_bits(&s->gb, 3);
4448
                    mode = rem_mode + (rem_mode >= mode);
4449
                }
4450

    
4451
                if(di==4)
4452
                    fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );
4453
                else
4454
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4455
            }
4456
            write_back_intra_pred_mode(h);
4457
            if( check_intra4x4_pred_mode(h) < 0)
4458
                return -1;
4459
        }else{
4460
            h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4461
            if(h->intra16x16_pred_mode < 0)
4462
                return -1;
4463
        }
4464
        if(CHROMA){
4465
            pred_mode= check_intra_pred_mode(h, get_ue_golomb_31(&s->gb));
4466
            if(pred_mode < 0)
4467
                return -1;
4468
            h->chroma_pred_mode= pred_mode;